section 1 bio-medical / bioengineering - uab...airborne infection is considered to be a major factor...

SECTION 1

BIO-MEDICAL / BIOENGINEERING

ASME 2012 Early Career Technical Journal - Vol. 11 1

ASME Early Career Technical Journal 2012 ASME Early Career Technical Conference, ASME ECTC

November 2 – 3, Atlanta, Georgia USA

STUDY OF TUBERCULOSIS SPREAD IN CLINICAL SETTINGS IN RESOURCE-CONSTRAINED COUNTRIES USING COMPUTATIONAL FLUID DYNAMICS (CFD)

Adeel Khalid, Christina Scherrer Southern Polytechnic State University

Marietta, GA, USA

ABSTRACT Tuberculosis is an infectious disease that can spread through air from a person’s cough or sneeze. Developing countries do not have the resources to take full preventive measures to contain the spread of tuberculosis (TB) through airborne particles. An infected patient visiting a health clinic or a local hospital is likely to spread the disease to other non-infected patients and healthcare workers. In this study, the goal is to look at the current state of affairs in terms of clinical patient care, patient isolation, building design, air circulation etc. in resource constrained countries, and investigate and suggest improvements in these areas. The data for this study is obtained from Centers for Disease Control (CDC) and published literature.

The spread of TB droplets can be modeled using the Lagrangian transport model of fluid flow. For this purpose, Computational Fluid Dynamics (CFD) is used. CFD simulations based on multi-physics approach can provide a good predictive efficacy of velocity flow field and particle concentration distribution in a room. Achieving accurate experimental measurements in hospital settings poses practical difficulties and is subject to high costs. In this research the CFD simulation is performed using a commercial software FLUENT. The boundary conditions are validated against verified data for comparative analysis [1].

INTRODUCTION Airborne infection is considered to be a major factor in the

spread of general infectious respiratory diseases, such as tuberculosis [2]. Airborne infection is caused by polluted air that becomes mixed with the aerosols expelled through breathing, coughing, or sneezing by an infected patient. Many clinical settings, including hospitals in resource-constrained countries, have occupants positioned closely for several hours and with a low air exchange rate, resulting in a high risk of disease transmission. To be able to determine the ideal room conditions to minimize the risk of airborne disease spread in that environment, it is important to study the transport of expiratory droplets and identify the zones under high risk. Uniform mixing of particle concentration is typically the assumption in the medical literature, but is not adequate to precisely determine the risk of disease transmission [3,4].

While best-practices exist for transmission reduction [5], these are cost-prohibitive in resource-constrained countries. There is growing evidence [6] indicating that infected airborne particles may play a greater role in the spread of infection than previously perceived. The objective of this on-going study is to determine cost-effective solutions to contain the hospital and clinic-based spread of TB in such countries. For that reason, we consider simple interventions such as open window locations, patient segregation, exhaust fans and screens or partitions in addition to more expensive options such as mechanical ventilation and UV lights [7].

CFD models are widely used to predict contaminant transport in indoor environments [8-13]. CFD is a flow modeling technique that solves Navier-Stokes equations to determine the motion of fluid substances. Incompressible flow of Newtonian fluids is governed by the following Navier Stokes equation.

𝜌 �𝜕𝑣

𝜕𝑡+ 𝑣 ∙ ∇𝑣� = −∇𝑝 + 𝜇∇2𝑣 + 𝑓 (1)

As indicated by the generalized Navier Stokes equation,

the flow field velocity v is determined by the pressure differential ∇𝑝, and the viscous forces μ are taken into account. For long term steady state, the 𝜕𝑣

𝜕𝑡 term is ignored. The

gravitational body force f on the fluid particles is also ignored due to the particle’s small size (<0.3microns).

In this study, the infected patient is modeled as a source of droplet generation. The spread of droplets through the room is modeled using the airflow rate or the number of air changes over a given amount of time. Using CFD, we determine the concentration of droplets at any given point in the room. Since the flow in a patient room is very complicated, the CFD model must be validated in order to produce reliable results. It is important to have high quality data with sufficient information on boundary conditions to validate the CFD models. Once established, the model can be expanded to solve more complex problems.

VALIDATION STUDY

There are many difficulties in carrying out accurate

experiments that analyze the behavior of airborne disease


particles. The purpose of this study is to systematically analyze the airborne particle behavior by numerical simulations. Banks et al. claim that validation examples, which would bolster the credibility of CFD studies, are seldom provided [10]. Without evidence that a CFD technique can reproduce experimental results in a similar situation, it is hard to have much confidence in the results. Therefore in this study the CFD model is validated against published results. Murakami et al. [1] measured the airflow field and particle concentration distributions in two ventilation systems under isothermal conditions. Since Murakami’s research provided all the necessary experimental details, in order to validate the accuracy of the simulations, the authors compared the numerical simulations with Murakami et al.’s published experimental results [1]. Zhang et al. [14] have also used Murakami’s work for validation purposes.

A high forced convectional flow-type room with particle injection is modeled. Numerical simulation is performed in ANSYS Fluent, based on the standard k- є (Reynolds-Stress) two equation model, coupled with Discrete Phase Model (DPM) for particle injection. After a numerical solution has been obtained, the results are post-processed for infection concentration visualization. The dispersion of TB droplets submerged in air is modeled using Lagragian, Discrete Phase Model (DPM) mass transfer. Adiga et al. showed that for high convective flow, the DPM model closely predicts the droplet transport timescale and agrees with the experimental results [15]. Shimada et al. showed that the changes in the concentration with time and the dependence of the concentration distribution on the position of the particle source are well reproduced by numerical calculations [16].

The Lagrangian discrete phase uses the Euler-Lagrange approach. The fluid phase is treated as a continuum by solving the time averaged Navier-Stokes equations, while the dispersed phase is solved by tracking a large number of droplets through the calculated flow field. The dispersed phase can exchange momentum, mass and energy with the fluid phase. The droplet trajectories are computed individually at specified intervals during the fluid phase calculations. This model is appropriate for simulating high momentum spray [15].

The configuration and specifications of the model room used for validation simulation are shown in Figure 1. The room dimensions are L x W x H = 3.29m x 5.85m x 2.8m. The isothermal three dimensional properties are analyzed. The supply velocity through the inlet vents is 1m/s. Supply inlet dimensions are 0.57m x 0.57m. The exhaust outlet dimensions are 0.64m x 0.25m. Water droplet liquid particles are generated by a plain orifice atomizer. The particle flow rate is 0.05kg/s. The injector inner diameter, which determines the size of the injected particles, is set at 3 microns.

Figure 1: Model Room used for simulation validation An unstructured tetrahedral mesh is generated to best fit

the room shape. The mesh size is set to 0.1m near the inlet and outlet vents. The overall mesh is finer than the mesh generated in the Murakami study, which is expected to generate more detailed results.

The expiratory droplets are tracked using the Lagrangian approach. It is assumed that the droplets exhaled during breathing are of one size. A steady state solution is first obtained and used as the initial condition for the expiratory droplet transport case. The particle generation rate is assumed to be constant. The boundary conditions for exhalation are as follows [17]:

• Flow rate and direction • Mouth or nose opening area • Temperature • Size distribution of the droplets These parameters can have considerable variation among

people. The CFD simulations by Chen et al. have treated these boundary conditions as constant flow rate with assumed direction, temperature, and appropriate climate for the area of the mouth opening [17, 18]. Accurate boundary conditions are important for precise prediction of droplet transmission. There is substantial literature on cough droplet size distribution, exhaled air temperature, and the flow dynamics of a cough i.e. flow rate, flow direction, and the area of the mouth opening during a coughing process [17]. Many experimental studies on coughing have shown that the average velocity of coughing varies from 1 to 10m/s [13]. The average velocity of breathing is slow, so it can be treated as a stationary source. In this study, only the transport of contaminants exhaled by coughing is modeled.

Patients often occupy a bed in an in-patient room using different postures, such as lying supine or sitting. Transport and dispersion of particles of different sizes exhaled by a seated or supine person on a bed in a room that has mixed ventilation have been investigated [13]. The simulation results show that the coughing direction has an obvious effect on the dispersion of different size particles in an unsteady condition. In this study, the patient is modeled lying in a supine position. The particle source injector height is 0.8 m. The boundary conditions used in this study are listed in Table 1.


2a Velocity vectors section view 2b Velocity vectors section view

2c Velocity vector plan view 2d Velocity vector plan view

2e Velocity vector section side view 2f Velocity vector section side

view Figure 2: Comparison of Velocity Flow Field

Results from our study (at left) versus Murakami (at right)

Table 1: Boundary Conditions Supply Vents (x2)

Velocity = 1m/s Turbulence Kinetic Energy = 1 m2/s2 Turbulence Dissipation Rate = 1 m2/s3 Size = 0.57m x 0.57m

Exhaust Vents (x4)

Ambient Pressure = 0 Pa Backflow Turbulent Kinetic Energy = 1 m2/s2 Backflow Turbulent Dissipation Rate 1 m2/s3 Size = 0.64m x 0.25m

Wall Boundary

Stationary Wall No Slip Condition Roughness Height = 0m Roughness constant = 0.5

Human Mouth

Mouth Opening for coughing Continuous Flow rate = 0.05 Kg/s Mouth Area = 0.0134 m2 Turbulent Dispersion – Discrete Random Walk Random Eddy Lifetime Plain orifice atomizer Number of particle streams = 60 Particle type – Inert Particle Size = 3.1x10-7

Figure 2 shows a comparison between Murakami’s

results and those obtained in this study. Velocity flow fields determined from Fluent are shown in Figures 2a, 2c, and 2e. These figures are comparable to the results obtained by

Murakami et al. [1] as shown in corresponding figures 2b, 2d, and 2f. These preliminary results illustrate that the numerical model is feasible for modeling indoor particle concentration distribution. The velocity flow field is shown as a contour in Figure 3. The velocity flow field is more discernible in the 3 dimensional volume rendering shown in Figure 4. The contour and volume rendering are not available in the original study and are provided here simply to better visualize the flow fields.

Figure 3: Velocity field contours

Figure 4: Velocity field volume rendering

Figure 5 shows the results of DPM concentration

simulation comparison. The airborne particles of less than 10 micro meters in diameter have approximately the same diffusion properties as a gas of the same density as air as far as time averaged space concentration distribution is concerned. The effect of gravity is negligible for particles of this size [1].

As can be seen from the figures, the velocity field and concentration results of the Fluent based CFD simulations match well with those of the published results. This provides us with the basis for validating the results. Simulation results highlight that the flow field and velocity distribution induced by the air inlets on the ceiling combined with the air exhaust vents near the floors produce wide circulation zones in the room and partial stagnation areas near the wall centers. The air flow also helps in lowering the particle concentration in the room. It reduces the spreading distance. With our validated


model, we can now use similar CFD models and boundary conditions to solve more complex problems.

a.

b. Figure 5: Comparison concentration contours. Case (a)

is from our study and case (b) from Murakami [1].

ROOM CONFIGURATION STUDY A full scale isolation room is considered in this study, with

the objective of comparing the impact of vent locations on contaminant spread. The solid model of the room is shown in Figure 6. The ventilation is provided from a typical HVAC primary air system which provides 12.5 air changes per hour (ACH) [19]. In the initial layout, the room is ventilated by a high induction air diffuser located in the center of the wall above the patient bed. The exhaust air is expelled by the return air diffuser located near the floor across the room. The exhaust vent of the ventilation system is set at a constant outlet pressure within the room. To study a steady state isothermal flow of air, no other methods of air exchanges are modeled (e.g. doors, bathroom, appliances or other items in the room). Air only passes through one inlet and one outlet. A single large room is modeled to reflect a typical waiting room in a developing central African region. The room is 6m x 6m x 4m, inlet vent 0.5m x 0.5m, and outlet vent 0.5m x 0.5m. The bed (1m x 1m x 2m) is located in the center of the room against the wall. A single patient is lying in a supine position. The head of the patient is modeled using a solid cylindrical geometry. For simplification, coughing or sneezing is modeled as a constant flow-rate through the mouth, an outflow surface on the head. Infectious droplets are 0.31μm.

To investigate the dispersion of fluid particles injected from coughing and sneezing, a steady state simulation combined with incompressible Navier Stokes equations coupled with Discrete Phase Modeling (DPM) is performed. In the areas of air flow, i.e. the inlet and exhaust vents and the

mouth, a fine mesh is employed, while a course mesh is used elsewhere. This is selected to strike a balance between accuracy and convergence time. A mesh of 28,633 tetrahedral elements is obtained for an un-structured grid. The viscous Reynolds – Stress model coupled with DPM interaction is used for simulating the cough and sneeze. No-slip conditions are applied to the wall. The model is treated as isothermal. Gravity is again neglected in the room, due to the small particle size. The droplets are assumed to enter the room at a constant rate through a small area located on the top of the bed, representing a point source of an infectious patient.

Figure 6: 3D Model of the Isolation Room The model is solved using first order discretization to

find steady state simulations. The boundary conditions, including the injection, are similar to the conditions used in the validation study. At the outlet, the mass flow boundaries are specified to ensure that the mass flow rate out of the flow domain corresponds to the flow rate into the domain. Wall functions are applied in conjunction with the RNG k-ε turbulence model. The walls are stationary and adiabatic. During particle dispersion simulation, particles that exit the room subsequently terminate their trajectories. The volume rendering of velocity flow fields of ventilation and diffusion air patterns inside the room are shown in Figure 7.

The flow field and velocity distribution induced by the air vents combined with the exhaust vent produce wide recirculation zones in the room. It also creates some partial stagnation areas near the ceiling, bed and walls. The airflow currents indicate the mixed convection in the cross section shown in Figure 7. Variable direction of the velocity flow field provides widespread and homogeneous distribution of air in the room. The flow is approximately symmetrical about the longitudinal centerline. The airflow current near the top of the room are stronger than those near the floor. Airflow near the floor is partially obstructed due to the presence of the bed. The airflow patterns are similar to those obtained by Murakami et al. [1].


Figure 7: Volume rendering of velocity field

The droplets are continuously exhaled with velocity of

coughing. The small droplets initially follow the coughing stream and then the bulk airflow in the room. The mean air velocity value of the central diffuser, calculated using simulation results, is 1.87 m/s. The airflow is laminar. In the areas near the outlet, the air velocity increases. The average air velocity near the inlet is 1m/s and the average air velocity near the outlet is 3.9m/s. This is due to the venturi effect. The area of the outlet vent is smaller than that of the inlet vent. The velocity flow field in the bed plane is shown in Figure 8. The velocity flow field in the vertical plane parallel to the position of the patient mouth is shown in Figure 9.

Figure 8: Velocity flow field in the bed plane (Top View)

The flow near the patient’s mouth interacts with the inlet airflow, resulting in an up-flow draft. This helps the light particles from the patient’s mouth to get caught in the circulating air and eventually being removed from the exhaust vent. The larger particles (greater than 10 micro meters) are heavy. They do not get transported through the air. This is shown in the concentration/velocity contour/vector plot on a vertical plane through the center of the room in Figure 9.

In the area near the bed, high levels of concentration of the fluid particles ejected from the patient’s mouth are localized in the vicinity of the patient’s mouth. This implies that the particle-droplet transport distance generated by coughing or sneezing of a patient is limited. The contaminant distribution is for the steady state with a continuous contaminant release from

the mouth of the patient. The contaminant concentration is low near the lower parts of the room, which is typical of displacement ventilation [20].

Figure 9: Velocity flow field in the vertical plane parallel to

patient’s mouth

When the small contaminant particles are trapped in the wake, they eventually get removed from the room with the air through the exhaust. The local air velocity near the coughing patient’s mouth is at or below the expected value of 10m/s. The maximum concentration contamination in the room is within 0.2 m distance from the patient mouth. The maximum concentration value is 1.54 kg/m3. Although there is a high particle concentration near the source location, little contamination is distributed across the rest of the plotted plane. The droplet distribution is highest in the vertical direction. This can be attributed to the updraft near the patient’s mouth. The droplets also get distributed in all directions around the patient’s mouth. However the droplet concentration sharply decreases as the distance from the patient’s mouth increases. Beggs et al. [21] highlight the complexity of airflow in rooms and its dependence on the local room design. They suggest that the forced ventilation method may not be well suited to hospital rooms, and convention ventilation may better control the spread of infection. Simulation results indicate that the ventilation system within a single room can have a significant effect on the distribution of airborne infectious material, and thus the risk of cross-infection.

Next the location of the inlet and outlet vents is varied, and the impact on the concentration of infectious particles in the room is studied. Specifically, we first visually map the concentrations in the room. Then, we numerically study the concentration of infectious particles at three locations that a healthcare provider would likely assist a patient.

In the following figure, the scale is the same for each of the concentration mappings. Everything else is the same as the base case above, except for the placement of the inlet and outlet vents. Solid models of the layouts are provided next to the concentration maps. Case (a) is the base case, with outlet and inlet vents both near the ceiling. In case (b), the outlet vents are moved close to floor level. In case (c) inlet vents are


at floor level and outlet at ceiling level. In case (d) inlet and outlet vents are all near floor level. In case (e) all vents are on the ceiling. In case (f) inlet vents are on the ceiling and outlet vents are near floor level. In case (g) inlet vents are close to floor level and outlet vents are on the ceiling. In case (h) inlet vents are on the ceiling and outlet vents are close to ceiling level. Finally, in case (i) inlet vents are close to ceiling level and outlet vents are in the ceiling.

a.

b.

c.

d.

e.

f.

g.

h.

i. Figure 10: Solid model and concentration map in the

standing plane (Top View) for cases a-i

The concentration mappings at standing height are fairly similar, yet there are some notable differences. As expected, there is an area of increased concentration at the patient’s head, though it is fairly small and surrounded by an area of reduced concentration. All exhibit some area of increased concentration near the sides of the room and very little concentration in the area immediately to the sides of the bed and across the room. In the first four cases, there is a higher pocket of concentration below the foot of the bed. Case (d), which represented the inlet and outlet vents both near floor level, has the highest concentration overall. This can be attributed to the velocity undercurrents that prevent uniform mixing of air throughout the room. Configuration (d) may also

Particle Mass Concentration Contour Legend


cause the contaminants to linger the room for an extended period of time, thus increasing the infection risk. Case (f), which represented inlet vents in the ceiling and outlet vents near floor level, had the lowest concentration of infectious particles.

The concentration maps at sitting height were also studied and there were no significant differences. (Those maps are available from the authors upon request.) Finally, the concentration of infectious particles are numerically studied at locations that healthcare workers are most likely to be at to assist patients – standing at the side of the bed, standing at the foot of the bed, and sitting in a chair at the side of the bed

Table 2: Contaminant Concentration values (Kg/m3)

Foot of bed (Standing)

Bedside (Standing)

Bedside (Sitting)

x=3, y=1.5, z=-2.25

x=3.75, y=1.5, z=-0.25

x=3.75, y=1.2, z=-0.25

a 1.46E-06 1.01E-08 2.65E-09 b 1.52E-06 1.65E-08 1.44E-08 c 1.46E-06 0.00E+00 0.00E+00 d 1.65E-06 0.00E+00 0.00E+00 e 1.43E-06 0.00E+00 0.00E+00 f 8.09E-07 0.00E+00 0.00E+00 g 1.23E-06 0.00E+00 0.00E+00 h 1.50E-06 0.00E+00 0.00E+00 i 1.48E-06 2.54E-09 0.00E+00

These results show that across the various layouts the highest risk location is clearly the foot of the bed. There is little significant difference in infectious concentration standing versus sitting at the side of the bed. Note that this is affected by the assumptions of cough direction and other settings used in this model, but still provides a useful insight for health practitioners. In addition, layout (f) which represented inlet vents in the ceiling and outlet vents near floor level had the lowest concentration of infectious particles at the three locations. Layouts (a) and (b) – inlet vents close the ceiling level and outlet vents at near the floor or near the ceiling - had the highest concentrations. CONCLUSIONS AND DISCUSSION

This research is a first step toward insights for determining optimal design of a patient room in a clinic or a small hospital in a resource constrained country. The room reconfiguration will help minimize the risk of air-borne disease (e.g. TB) transfer from an infected patient to other patients or hospital staff. Air flow patterns are observed using CFD models. CFD modeling is a powerful tool for investigating ventilation strategies. The cough or sneeze is modeled at a

steady state condition in an isothermal room. It is observed that the location and the speed of airflow can affect the transport of infected particles through the air. If the infected patient is placed such that the droplets either get transported out of the room through the exhaust air or remain localized, then the spread of the disease can be minimized. It is determined that large droplets (> 10 micro meters) do not get transported and remain localized near the source.

The results of this study indicate that the position of air-supply inlets on the walls near the ceiling and the exhaust vents at the opposite side of the room provide a circulation motion of air. It also provides an up-draft near the patient which helps transport contagious droplets to the exhaust. If on the other hand, the exhaust vents are located on the wall behind the patient’s bed, the coughed particles can get trapped in the region around the bed. This can minimize the coughed droplet diffusion and hence increase the risk of infection transport. The high induction air inlet diffuser and the exhaust vents provide an effective means of controlling droplet flows containing contaminants. This type of analysis helps in predicting air recirculation zones. The analysis of velocity field and concentration mapping is effective in identifying examples of such zones in the room. Current simulation results yield meaningful findings and recommendations for disease control and careful design and optimization of ventilation in hospitals and clinics for preventing spread of airborne diseases. This work can provides key insights for hospital management to control the room design (dimensions, arrangement etc.), position and dimensions of openings (doors, windows, window orientation), and in particular, type and location of supply and return air diffusers and air velocity. FUTURE WORK

The results obtained in this study will be significantly expanded. The studies will include different room configurations, number of patients in the room, their location, and types of ventilation (natural, mechanical of various levels) and room layouts. They will also include the effects of various interventions, such as exhaust fans, partitions for segregating patients, and UV lights. This would help determine the optimum setting of the room in terms of various variable settings that will help minimize the spread of airborne contagious diseases. The verified and validated CFD results could be expanded to include other scenarios for which experimental data does not exist. This technique can be further extended to include more detailed models of the physics of contamination transport – for example, the temperature and relative humidity of the air, to account for buoyancy of particles and for changes in particle size due to evaporation.

The Wells-Riley equation for airborne infection is used to estimate the infection risk given the particle concentration in the room [3,22]. Researchers often assume complete air mixing leading to uniform concentration of aerosols throughout the space, resulting in the quantification of the average risk rather than the expected range. CFD simulations provide a far better estimate of the particle concentration distribution in rooms of


various sizes and geometries. Our goal is to extend the present study, and combine it with the Wells-Riley model to estimate the individual’s cumulative infection risk [22]. We will also combine this with cost data to provide clinic designers and public health officials with the information to make cost-effective decisions about their investments. ACKNOWLEDGEMENTS This work was partially supported by the National Science Foundation under grant number BRIGE-0927095. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. The authors would also like to thank Michele Pearson, MD from the Centers for Disease Control for her insights into the project. REFERENCES: [1] Murakami, S., Kato, S., Nagano, S., Tanaka, Y., ‘Diffusion characteristics of airborne particles with gravitational settling in a convection-dominant indoor flow field,’ 1992, ASHRAE Transactions, Vol. 98, part 1, pp. 82-97 [2] Zhu, S., Kato, S., Yang, J-h., ‘Investigation into airborne transport characteristics of airflow due to coughing in a stagnant indoor environment,’ 2006, ASHRAE Transaction, Vol. 112, Part 1, pp. 123-133 [3] Noakes, C. J., Sleigh, P. A., ‘Applying the Wells-Riley equation to the risk of airborne infection in hospital environments: The importance of stochastic and proximity effects,’ 2008, Indoor Air Conference, Copenhagen, Denmark, Paper ID: 42 [4] Beggs, C.B., Noakes, E. J., Sleigh, P. A., Fletcher, L. A., Siddiqi, K., ‘Transmission of Tuberculosis in confined spaces: an analytical review of alternative epidemiological models,’ 2003, International Journal of Tuberculosis and Lung Disease, Vol. 7, Issue 11, pp. 1015-1026 [5] World Health Organization, 2009, Stop TB Department. WHO Policy on TB Infection Control in Health Care Facilities, Congregate Settings, and Households. WHO Press: Geneva, Switzerland. [6] Beggs, C. B., Kerr, K. G., Noakes, C. J., Hathway, E. A., Sleigh, P. A., ‘The ventilation of multiple bed hospital wards: review and analysis,’ 2008, American Journal of Infection Control, Vol. 36, Issue 4, pp. 250-259 [7] Boyce, P., et al., ‘Controlling Tuberculosis transmission with ultraviolet irradiation,’ 2003, Technical report Rensselaer Polytechnic Institute, Lighting Research Center [8] Ko, G., Thompson, K. M., Nardell, E. A., ‘Estimation of Tuberculosis risk on a commercial airliner,’ Risk Analysis, Vol. 24, No. 2, 2004 [9] Gupta, J. K., Lin, C, H., Chen, Q., ‘Prediction of spatial and temporal distribution of expiratory droplets in an

aircraft cabin,’ International High Performance Building Conference at Purdue, July 12-15, 2010 [10] Banks, D., Meroney, R. N., Petersen, R. L., Carter, J. J., ‘Evaluation of Fluent for predicting concentrations on buildings,’ 2003, Air and Waste Management Association Annual Conference, San Diego, CA. pp. 11. [11] Beggs, C. B., Kerr, K. G., Noakes, C. J., Hathway, E. A., Sleigh, P. A., ‘The ventilation of multiple bed hospital wards: review and analysis,’ 2008, American Journal of Infection Control, Vol. 36, Issue 4, pp. 250-259 [12] Mazumdar, S., Yin, Y., Guity, A., Marmion, P., Gulick, B., Chen, Q., ‘Impact of moving objects on contaminant concentration distribution in an impatient ward with displacement ventilation,’ 2010, HVAC and Research, 16(5), pp. 545-564 [13] Yin, Y., Gupta, J. K., Zhang, X., Liu, J., and Chen, Q., ‘Distribution of contaminant exhaled out by a patient with different postures and exhaling modes in a single bed patient room,’ Buildings and Environment, 46(1), pp. 75-81 [14] Zhang, Z., Chen, Q., ‘Experimental measurements and numerical simulations of particle transport and distribution in ventilated rooms,’ 2006, Atmospheric Environment, 40 (18), pp. 3396-3408 [15] Adiga, K. C., Sheinson, R. S., William, F. W., Ayers, S., ‘Modeling ultra fine mist transport and its implications on fire suppression behavior,’ Suppression and detection research and application – a technical working conference (SUPDET 2007) [16] Shimada, M., Okuyama, K., Okazaki, S., Asai, T., Matsukuru, M., Ishizu, Y., ‘Numerical simulation and experiment on the transport of fine particles in a ventilated room,’ 1996, Aerosol Science and Technology, 25:3, pp. 242-255 [17] Gupta, J. K., Lin. C. H., Chen, Q, ‘Flow dynamics and characterization of a cough,’ 2009, Indoor Air, 19, pp. 517-525 [18] Eames, I., Shoaib, D., Klettner, C. A., Taban, V., ‘Movement of airborne contaminants in a hospital isolation room,’ 2009, Journal of the Royal Society Interface, 6 S757-S766 [19] Balocco, C., Lio, P., ‘Modeling infection spreading control in a hospital isolation room,’ 2010, Journal of biomedical science and engineering, 3, pp. 653-663 [20] Mazumdar, S., Yin, Y., Guity, A., Marmion, P., Gulick, B., Chen, Q., ‘Impact of moving objects on contaminant concentration distribution in an impatient ward with displacement ventilation,’ 2010, HVAC and Research, 16(5), pp. 545-564 [21] Beggs, C. B., Kerr, K. G., Noakes, C. J., Hathway, E. A., Sleigh, P. A., ‘The ventilation of multiple bed hospital wards: review and analysis,’ 2008, American Journal of Infection Control, Vol. 36, Issue 4, pp. 250-259 [22] Griffin, P., Minarcine, C., Scherrer, C., ‘Reducing TB incidence in clinical settings in resource constrained countries,’ 2011, Proceedings of the Industrial Engineering Research Conference



November 2 – 3, 2012, Atlanta, Georgia USA

DEVELOPMENT OF A REHABILITATIVE EXOSKELETAL ARM

Daniel Garcia, Victor Soto, Yojans Lurbe, Melissa Morris, Sabri Tosunoglu

Florida International University Department of Mechanical and Materials Engineering

Miami, Florida USA

ABSTRACT Reduced motor capacity of the upper extremities is a common result of strokes, spinal cord injuries, accidental injuries, and neurodegenerative diseases. Sensorimotor recovery can be attained through gradual and repetitive exercises. In recent years, robot-assisted rehabilitation has been shown to improve treatment outcomes in these cases. This paper aims to discuss a potential method of rehabilitation through the use of a robotic exoskeletal device that is designed to conform to the shape of an arm. Three different program methods were developed as modes of exercise and therapy to achieve passive exercise, assisted motions, and resistive-active exercise.

INTRODUCTION Reduced motor capacity of the upper extremities is a

common result of strokes, spinal cord injuries, accidental injuries, and neurodegenerative diseases. Sensorimotor recovery can be attained through gradual and repetitive exercises [3-5]. In recent years, robot-assisted rehabilitation has been shown to improve treatment outcomes in these cases [6-10].

Current exoskeleton rehabilitative devices have multiple advantages over traditionally manual techniques, including [2]:

Data tracking for performance feedback The ability to apply controlled forces at each joint as

well as magnitude adjustment of such forces based on patient needs

They can be adjusted for multiple limb sizes to fit different patients

They can replicate the majority of the patient’s upper limb healthy workspace, using multiple degrees of freedom.

This device contains additional advantages over current

devices. First of all it will be portable. It is going to address a

very specific task, which makes it more user friendly, and last but not least it has a simple and cost effective design.

This bicep & tricep therapeutic device will have three modes of operation: passive, assisted motions, and resistive-active. A linear actuator provides the necessary movement of the exoskeleton and a pair of force sensors tracks the response of the patient to the therapeutic session. The passive mode is for patients that have complete muscle atrophy. In this mode the actuator does all the work to emotionally stimulate the patient. The assisted motions mode offers the patient force amplification. This mode allows patients with weak upper limbs to perform everyday life tasks such as lifting, pushing, pulling, etc. In this mode, the speed of the actuator is directly proportional to the force applied by the user. If the patient applies a higher force, the actuator moves faster, and vice versa. In the resistive-active mode, the user must apply a load on the load sensor that surpasses a certain threshold. When the robot detects this, it moves the actuator at a speed that creates resistance for the user. In this mode, if the load applied by the user falls below the threshold, the actuator stops.

Conceptual Design

The conceptual design considered as part of this system is

Figure 1 - Motion Schematic


shown in figure 1. This basic 1 Degree of Freedom (DOF) system consists of two links connected by a revolute joint. A linear actuator adjusts the angular separation between the two links as shown by the sequence in the figure.

Theoretical Work

For a robotic system to achieve static force equilibrium the following must be true [1]:

= − = − ∗ (1) Where

is the required actuator torque is the Jacobian transpose of the system is the force system applied at the end-effector of the robot

The force system was decomposed into its X and Y

components as shown in equation 2 and 3. = sin (2)

= cos (3)

Using the length of the actuated link ( ), the location of the end-effector was obtained: = cos (4) = sin (5)

Equations 4 and 5 where then differentiated with respect to φ to obtain the translational g-functions:

= − sin (6) = cos (7)

Equation 1 and the values obtained from equations 2, 3, 6,

and 7 were used in matrix form to calculate the required actuator torque as shown in equation 8. = −[ ] ∗ [ ] (8)

The curve shown in Figure 2 was used to estimate the force

delivered by the linear actuator for the nominal speed of the robotic system.

Figure 2 - Linear Actuator Force vs Speed Curve At a speed of 0.7 inches per second, the linear actuator can

deliver a force (FL) of up to 175 lb. As shown in Figure 3, the component of this force perpendicular to the actuated link is given by FP, and the distance from the joint to the pivot point of the linear actuator is given by L1.

Figure 3 - Actuator Torque

FP and its moment arm L1 were used to calculate the actual

torque delivered by the actuator as shown in equations 9 and 10. = ∗ sin (9) = ∗ (10)

Assuming an angle of 30 degrees for Ф, and an angle of 14 degrees for α, the torque delivered by the actuator is approximately 23 lb-ft. According to the static equilibrium analysis performed in equations 1 through 8, this actuator torque can support an external load FH of up to 50 lb.

A dynamic analysis was also considered for this robotic system. The dynamic equation for an n-link robotic arm is defined as: [ ∗] = , + + (11)

Given the slow movements and low mass content of

rehabilitative exoskeletons, ω and α are negligible. As a result, equation 11 reduces to equation 1.


DESIGN Figure 4 displays the final iteration of the design. The two

links consisted of two Aluminum U-Channels. Two titanium hollow shafts were used for the joints. Steel shoulder bolts were used to attach both ends of the linear actuator to each link. The linear actuator consisted of a 24v DC servo and a power screw. Two 10 lb, high resolution load cells were used on either side of the wrist brace to detect and measure the user’s actions. A pair of limit switches were employed for limiting the range of motion of the links to avoid damage to the actuator and for the safety of the users. To control and synchronize the motion of the actuator to the user’s commands, an Arduino Uno Microcontroller board and Pololu motor shield were employed. A complete list of all components used is listed in table 1.

Figure 4 - Major Components

Table 1 - Bill of Materials

Qty Component

1 6061 – T6 5” Aluminum U-Channel

1 6061 – T6 6” Aluminum U-Channel

2 6A1 – 4V Titanium Pivot Pin

2 Steel Shoulder Bolt ¼ in x 1 in

1 24V Linear Actuator 90lbf

1 DC Servo

2 10LBF Load Cell

2 Limit Switch

1 Arduino Uno Microcontroller Board

1 Pololu Motor Shield

PROGRAMMING The initial phase of programming began with

understanding the syntax of the wiring language that the Arduino uses. Wiring is a custom language that is derived from C++ with a few modifications for the simplification of syntax. Because the motor shield had already been decided upon as the main device to control the servos, additional libraries that were

written for the Arduino compiler had to be implemented and thus new syntax had to be added and understood.

The first iteration of the program began as a simple test of all the components to have a basic understanding of how the Arduino interacted with the device, as well as the response time and calibration of the sensors that were being used to detect force applied. Several tests were conducted with the first iteration which included: sensor detection, switch detection, servo initiation, as well as button mode switching.

After the initial trial, the code was rewritten to incorporate the intended design and flow chart (figure 5) that was created to execute the desired function of the exoskeleton robotic arm. The program begins with the initiation of all the global variables, along with the setup of the custom library, and pin modes. Several subroutines were developed that could be called upon on every loop of the program to read the sensors and limit switches. Once the program checks the sensor and switch values, it responds to the button mode by entering different conditional loops that control which desired function the arm executes. Divided into three conditions, the first program runs the cyclic flex and relaxation motion of the arm, the second program implements a linear function that acts as an assistive ramp up mechanism when sensor input is detected, and the third condition implements a linear function that slowly activates the servo to simulate a resistive force in response to sensor input.

Figure 5 - Program Logic


CIRCUITRY The motor shield occupied a majority of the digital pins

because of the PWM capabilities, the remaining digital pins were used for the switch inputs which all had drop-down resistors for functionality. The sensors were connected directly to the power source and ground, with the signal output connected directly to the analog input pins through the motor shield. Because the motor shield handled the inputs and outputs when interfaced with the Arduino, the remaining ports were just for connecting the power and motor to the shield directly.

PROTOTYPE CONSTRUCTION The prototype consists of a two piece extruded Aluminum

U-channel frame that was machined on a vertical mill and sanded down to eliminate sharp edges. Two hollow Titanium shafts were machined on a lathe and secured the two segments of the Aluminum channels with external retaining rings, as shown in figure 6.

Figure 6 - Frame Assembly

Initial testing of the linear actuator resulted in the motor

burning out from stalling when the piston accidentally reached its limit. As a result, a replacement motor of similar size and torque but with a much higher max RPM was obtained. The replacement motor operated at too high a speed, so a significant gear reduction was needed in order for the piston to move at a manageable speed. New timing belt pulleys and a longer timing belt were ordered from McMaster to replace the existing ones, changing the gear ratio from 3:2 to 5:1 before the change from rotation to linear motion. Active idler pulleys were machined out of Delrin in order to ensure that the belt wrapped around approximately 50% of the smaller driving pinion and prevent belt slippage and damage. This configuration is shown in figure 7.

Figure 7 - Linear Actuator Drive

With the linear actuator up and running again, two solid

blocks of Aluminum were machined on a CNC machine to create the joints between the linear actuator and the frame assembly. Soon after the adjusting sensor mount/wrist support was machined on a vertical mill and assembled onto the frame as shown in figure 8. At this point all the machined components were then sandblasted to provide a clean finish (figure 9).

Figure 8 - Linear Actuator Mounted


Figure 9 - Frame Assembly Sandblasted

Once the machined components were assembled, the

Arduino motherboard, along with the Pololu motor driver, and hand-built circuit on a breadboard were connected to the mechanism as shown in figures 10 and 11. This allowed testing of the software to ensure proper functionality of the mechanism.

Figure 10 - Electronics Connected

Figure 11 - Electronic Connections

Once the software was running properly, the electronic

boards were mounted and all the electrical connections were soldered on and routed neatly. Enough slack was left to ensure proper routing throughout the mechanism’s range of motion. Conforming arm rests that were printed with an SLS machine were then assembled to the mechanism, and Velcro straps were cut to length and attached (figures 12 and 13).

Figure 12 - Final Assembly Right Side


Figure 13 - Final Assembly Left Side

Figure 14 - User Demonstration

Testing and Test Results Testing of the mechanism consisted of confirming that

each sensor and actuator worked individually, before the assembly as a whole was tested. Both touch sensors were measured with a voltmeter to ensure that the circuit was normally open when untouched, and closed when the lever depressed the buttons. The load cells were measured with a voltmeter as well to ensure that there was a consistent change in voltage between the signal and ground wires when 5V was applied. The 24V motor was run independent of the linear actuator to confirm proper functionality, and then it was actuated when assembled to the linear actuator to ensure proper function of the pulley system and screw/nut mechanism.

Once all individual components were verified, the mechanism was assembled, with the exception of the linear actuator, and the early revisions of the software was run to ensure the logic was sound and proper movement of the actuator per load cell and touch sensor inputs. The final stage of testing prior to final assembly was verification of the linear

actuator reaction to load cell inputs from the user – the higher the load on the cell, the faster the linear actuator moved.

Once the mechanism was fully assembled and all wires were routed, final testing was performed. A user would strap their arm into the mechanism and operate the mechanism at various speeds and intensities to ensure proper functionality, reaction time, and calibrated for a comfortable input threshold for motor reaction (start-up force and variable speed control). This testing was performed for all three states of the mechanism: passive, assisted motions, and resistive-active.

The system performed well, with a low force threshold and good sensitivity in the assisted motions mode. A balance between the linear actuator force and speed was needed in the passive state in order to allow the mechanism to begin its motion up from a fully extended position (flexing), as it is the orientation that requires the largest amount of force from the actuator to begin movement after hitting the touch sensor. The resistive-active state resists motion well in order to make the user work for small slow movements of the mechanism, and may be altered via the software to allow for users of varying levels of strength.

Overall the mechanism performed quite well, but the ergonomics of the system are not the best. Velcro straps are used to attach the mechanism to the user’s arm as shown in figure 14, which provides a loose fit at best and is most noticeably lacking in the resistive-active mode. The mechanism as a whole has opportunities for weight reduction as having it strapped to your arm may cause shoulder soreness after extended periods of use. Finally, the size of the overall frame and packaging of the electronics also have the opportunity to be reduced for a more comfortable fit.

CONCLUSIONS AND FUTURE WORK This first generation robotic system contains the necessary

features needed for the device to be therapeutically sound. Further design improvements are needed however before clinical trials can be performed to ensure patient safety and reliability of the device. The fastening system for example needs to be improved, to ensure that the device is securely attached to a patient’s arm during the therapy sessions. Additionally, a soft padding material should be added to the parts that are in contact with the patient’s arm. An end-effector should also be added to the back side of the wrist brace to allow users to exploit the full load capacity of the robot.

REFERENCES [1] Tosunoglu, S, and D Tesar. Robotics & Automation. University of Texas at Austin. [2] Vertechy, Rocco. Development of a New Endoskeleton for Upper Limb Rehabilitation. Kyoto International Conference Center. June 2009. [3] Butefisch, H. Hummelsheim, P. Denzler, and KH Mauritz. Repetitive training of isolated movements improves the outcome of motor rehabilitation of the centrally paretic hand. J Neurol Sci, 130(1):59-68, 1995 [4] H.M. Feys, W.J. De Weerdt, B.E. Selz, G.A. Cox Steck, R. Spichiger, L.E. Vereeck, K.D. Putman, and G.A. Van Hoydonck. Effect of a Therapeutic Intervention for the


Hemiplegic Upper Limb in the Acute Phase After Stroke A Single-Blind, Randomized, Controlled Multicenter Trial, 1998 [5] J.H. van der Lee, R.C. Wagenaar, G.J. Lankhorst, T.W. Vogelaar, W.L. Deville, and L.M. Bouter. Forced Use of the Upper Extremity in Chronic Stroke PatientsResults From a Single-Blind Randomized Clinical Trial, 1999 [6] Reinkensmeyer, D.J., Kahn, L.E., Averbuch, M., McKenna-Cole, A.,Schmit, B.D., and Rymer, W.Z. 2000. Understanding and treating arm movement impairment after chronic brain injury: Progress with Arm Guide. Journal of Rehabilitation Research and Development, 37(6). [7] Jack D, Boian R, Merians A, Tremaine M, Burdea G, Adamovich S, Recce M, Poizner H. 2001. Virtual reality-enhanced stroke rehabilitation. Neural Systems and Rehabilitation Engineering, IEEE Transactions on [see also IEEE Trans. on Rehabilitation Engineering] 9(3):308-318.

[8] Cardoso L, daCosta R, Piovesana A, Costa M, Penna L, Crispin A, Carvalho J, Ferreira H, Lopes M, Brandao G, et al. 2006. Using Virtual Environments for Stroke Rehabilitation. Virtual Rehabilitation, 2006 International Workshop on pp. 1-5. [9] Stewart J, Yeh S, Jung Y, Yoon H, Whitford M, Chen S, Li L, McLaughlin M, Rizzo A, Winstein C. 2006. Pilot Trial Results from A Virtual Reality System Designed to Enhance Recovery of Skilled Arm and Hand Movements after Stroke. Virtual Rehabilitation, 2006 International Workshop on pp. 18-23. [10] A. Frisoli, A. Montagner, L. Borelli, F. Salsedo, M. Bergamasco. A force-feedback exoskeleton for upper limb rehabilitation in Virtual Reality. Applied Bionics and Biomechanics Vol. 00, No. 00, December 2008, 1-17.




FLUID DYNAMIC ANALYSISOF LUNG ALVEOLI UNDER MECHANICAL VENTILATION

CONDITIONS Trenicka K. RolleGraduate Student and Dr. Ramana PidapartiProfessor

Department of Mechanical and Nuclear Engineering Virginia Commonwealth University

Richmond, VA, USA

ABSTRACT The alveolar region of the lungs plays the most important

role in breathing, which is the process of gas exchange. It occurs between the alveolar membrane and the underlying capillaries. During mechanical ventilation, the distribution of forced air within lung parenchyma results in the overdistension of the alveolar wall, leading to a cascade of other conditions. These conditions include volutrauma/barotrauma (extreme stress/strain), atelectrauma (repeated opening and closing of collapsed alveoli) and biotrauma. If the aforementioned conditions are increased and worsen, multi-system organ failure (MSOF) will occur as a result. This study takes into consideration the computational fluid dynamics within the airways in the pulmonary acinus. Although many studies have been done on the alveoli in order to determine the stresses and strains induced, not many have taken into consideration the injurious effects of mechanical ventilation. In addition with each study the methods and models vary. The goal of this study is to investigate the fluid dynamics within the alveoli by comparing and contrasting the differences in various velocity inlet profiles.A finite element model was created and simulations were run using the ANSYS software package, specifically performing a computational fluid dynamics study using FLUENT. The fluid analysis will yield results that can be used to determine the wall shear stress, pressures and velocities on the alveolar wall whichcan be applied to a tissue model leading ultimately to a study on the microenvironment and how the mechanical properties of the region are affected.

1.INTRODUCTION The lung is one of the most important organs within the

body, having the responsibility of the most important function which is breathing. It consists of a network of branching bronchi, bronchioles, alveolar ducts and alveoli. The bronchiole trees are classified according to the term generations with a two ordering scheme.The trachea begins the tree at generation 0, whose average diameter and length are, respectively, D0 =1.8 cm and L0 = 12 cm in the healthy human adult, and ends in the terminal bronchioles [1]. Terminal bronchioles start at generation 16 and lead to respiratory bronchioles, the first generation to have alveoli which lead to alveolar ducts and finally alveoli sacs (generation 23) whose walls are entirely composed of alveoli [2]. Any particular changes to the physical

and mechanical properties of the lung parenchyma can prove to be fatal especially in the cases of many with severe respiratory diseases. According to a study done by the U.S Department of Health and Human Service, 61,987 adults ages 18 and over suffered from respiratory diseases including emphysema, asthma and chronic bronchitis out of a study of 229, 505 [3]. Many of these diseases deteriorate the lung, especially the region of the lower lung comprised of the alveoli. This is what is known as acute respiratory distress syndrome (ARDS) in which the mechanical properties of the lung parenchyma, such as lung compliance, are decreased [4]. Hence, lung disease is prevalent among many U.S. adults, and provided that these conditions become life threatening, mechanical ventilation is used as a tool to assist such patients by decreasing the work of breathing performed by the lungs. Although the intended purpose of mechanical ventilation is to assist with breathing and maintaining the life of the patient, in many cases it has resulted in the death of many patients and an exacerbation of preexisting conditions [5]. In addition healthy parts of the lung can be overextended, and these complications have been termed ventilator-induced lung injury (VILI). Another term associated with these ventilatory complications is ventilator associated lung injury (VALI). The mortality rate as a result of respiratory failure due to VALI is about 137 – 253 per 100,000 in the general population in the US alone [6].

The lung parenchyma consist of the respiratory bronchioles, alveoli and capillaries; these are the key elements of the lung that are essential to its function. The parenchyma is also considered to be a regulated mixture of alveolar walls and air. The cyclic application of physical stresses at the pleural surface of the lung results in length changes in the parenchymal structures yielding volume variations which result in applied strains at the alveolar walls. There have been many studies focused on the effects of mechanical ventilation on the diseased and healthy portions of the lung at the tissue and cellular level. Li et al. [14] studied the airflow analysis in the alveolar region, showing the influence of the geometric structure on the airflow field and pressure distributions. However this study gives no information on the mechanical forces induced by the interaction between the air and alveolar wall. A 2-D fluid structure analysis study by Dailey et al. [15] investigated how tissue mechanical properties and breathing patterns influence


deep-lung flow fields and particle dynamics. Nevertheless this study does not include mechanical ventilation.Using a strain energy function model and uni-axial tension tests on living precision-cut rat lung slices Rausch and Wall [16] as well as other colleagues were able to formulate a new material model for lung parenchymal tissue that allows for finite element simulations of the lung parenchyma. This study’s focus was to ascertain those material parameters for various stain energy functions (SEF) for the elastic model and then to combine various models of SEF’s to fit the experimental data.

In addition, Rausch and colleagues [17] investigated the local strain distribution by using a finite element simulation of X-ray tomographic microscopy scanned alveolar geometries. They were able to model the alveoli wall and attain strain distributions for single alveolar wall.

Nevertheless these studies have not introduced mechanical ventilation boundary conditions.Hence it was important to understand the fluid dynamics of lung alveoli under such conditions. Seeing that they are not fully understood, the goal was to perform a computational fluid dynamic study. Therefore this study is significant to discerning those mechanical forces that result in overdistension of the alveoli. Results can be used to help physicians establish safer ventilation rates to decrease patient mortality and present findings on where does injurious mechanical ventilation begin within alveolar tissue.

2. MATERIALS AND METHODS 2.1 3-D Model Geometry

There have been many studies on determining the shape of the alveoli, which is an important factor in the determination of pressure distribution, air flow pattern and particle deposition within these structures. In early studies the alveoli were illustrated as balloons with varying sizes, Clemente et al. [20]. However, many studies in recent times have modeled alveoli geometries as having a polygonal shapes of octahedrons, dodecahedrons spheroids etc., De Riuk et al. [21] modeled the alveoli as rhombic dodecahedron which is a polygon constructed with twelve rhombuses adjoined by fourteen vertices [7]. In this particular study we modeled the alveoli sac as a sphere; hence we assume a spherical geometry [10, 11]. A study performed by Harding Jr., and Robinson modeled the alveoli as sacs such that an image of the SEM of alveoli revealed that alveoli are present in a wide distribution of shapes, from spherical to multi-hedron [7a]. The dimensions of the model are in compliance with typical alveolar ducts and sacs in human lungs. The model has an inlet duct diameter of 200µm and an alveolar entrance length of 100 µm [8]. These dimensions are taken from Dailey, whose dimensions are consistent with the dimensions used by Haefeli-Bleuer & Weibel as well as Ochs [12, 13]. Figure 1 illustrates the model geometry for the solid and fluid components.

It is important to have an accurate mesh in order to obtain the best results as it is common to attain a volume error when

meshing spheres. The element size was set to 1e-5. The model once meshed had a total of 490011 fluid elements and 94858 nodes. Figure 2 illustrates the meshed fluid model.

Figure 1. Alveoli Model Geometry

Figure 2. Meshed Fluid Model

2.2 Governing Equations The analysis was solved using a transient analysis in

FLUENT [9], hence making the solution time dependent. The fluid flow was solved using the incompressible continuity and Navier Stokes equations.

ρδvi= 0, (1) δxi ρδvi + ρvjδvi = - δp + μ δ2vi , (2) δt δxjδxi δxj δxj ρ is the fluid density, µ is the fluid viscosity, vi is the

velocity vector, xi is the position vector, t is time, and pis the fluid pressure [8]. All fluid properties for air are taken at body temperature of 37 °C.

FLUENT employs a second order Euler backward method to solve equations 1 and 2.


2.3 Boundary Conditions and Flow Parameters No slip wall conditions were applied to the alveolar ducts

and walls. The simulation was characterized by an inlet velocity boundary condition. A user defined function (UDF) for the inlet was created and applied to the inlet of the model to define the flow. A UDF had to be defined because FLUENT, being the solver for this simulation, only allows for a UDF in order to define a unique or user specific input. The pressure throughout the model was set to 101 kPa. A transient analysis was utilized in which the total time would cover a breathing cycle of 4s. Specifically from 0 – 2 seconds is considered inspiration, on the other hand from 2 – 4 seconds is considered exhalation. The time step value was set to 0.01 with 20 iterations per time step to give the most accurate results possible. The calculated flow rates converge at a time step of 0.01 seconds [11].The convergence criteria for the residuals were set to 1E-4.

An inlet velocity profile was applied to the opening of the alveolar duct. The profile that represents the ramped waveform is the general waveform for mechanical ventilation [19].In terms of the other profiles the goal was to vary the profile in order to observe the differences in fluid parameters as compared to the typical ramped profile. The velocities at generations 21 on down are not known, and many studies have used various inlet waveforms, pressure inlet conditions and other user defined functions in order to determine flow within the alveolar sac. Assuming an idealized dichotomous lung,the flow at the alveolar ducts would be based on the tracheal flow rate in which the flow varies with each generation number. This can be derived from Q(n)=Q0/2

n [18]. Hence using a 60 L/min tracheal flow rate and converting that flow rate to a velocity, the velocity was determined for the alveolar duct at generation 22 and calculated at specific times. Figure 3 depicts the flow waveform of several inlet velocity profiles used in this analysis.

Figure 3a. Scaled tracheal (Ramped) flow waveform

Figure 3b. Sinusoidal Waveform

Figure 3c.Constant Flow at 1mm/sec

Figure 3d.Tracheal Flow Waveform (Turbulent)

3. RESULTS After running each simulation with the various flow

profiles, results were obtained. There have been several studies in which an airflow analysis or computational fluid dynamics (CFD) study has been performed in the alveolar region of the lungs. In a study performed by Li [14] the velocity results obtained had a maximum value of 1.2 and 0.7 m/s. In another study by Harding [11] the maximum flow obtained in the alveolar sac was 9.3E-4 m/s. The velocity values obtained in this study are closer to those obtained in that by Harding. However, it is expected for there to be differences, as the boundary conditions were not the same; neither was the model,


and geometry does affect how the flow field forms. In addition, the main difference would be that this study takes into consideration mechanical ventilation, whereas the aforementioned studies do not. Moreover there have been several CFD studies and airflow analyses of the alveolar region, but not many, if any at all, are with respect to mechanical ventilation. Hence, this is what makes this study unique, and in terms of justifying the results obtained, more studies would need to be performed taking into consideration mechanical ventilation in order to obtain some type of consistency in results.

Three specific parameters were considered in this study,which include velocity, wall shear and pressure. Figure 4 depicts the velocity flow fields for each of the inlet velocity profiles. Figure 5 illustrates the pressures distribution and magnitude and lastly Figure 6 shows the wall shear distribution and magnitude.

Figure 4a. Flow Field Results (Ramped Profile)

Figure 4b. Flow Field Results (Sinusoidal Profile)

Figure 4c. Flow Field Results (Constant Profile)

Figure 4d. Flow Field Results (Tracheal Flow)

Figure 5. Pressure Distribution (Constant Flow)


Figure 6. Wall Shear Distribution (Sinusoidal Profile)

There were noticeable differences in the flow fields for

each case. However for the scaled tracheal, constant and tracheal results for velocity the distribution of flow within the alveolar sacs are generally similar. The sinusoidal profile results for velocity show flow throughout the entire model, including alveolar ducts where some high flow regions exists. Figure 5 shows the pressure distribution results for the constant flow. The sac with the highest pressure can be found on the middle sac where the red contour depicts this particular region. We can assume that based on geometry and with the inlet velocity having only a y component initially, that once flow begins through the model the middle sac being attached to the alveolar duct that is not at an angle will have the highest flows, hence the highest pressures. Furthermore Figure 6, illustrating wall shear for the sinusoidal profile, shows that the regions where the highest wall shear can be found are in the corners where the ducts begin to branch off and at the beginning of each alveolar sac. Table 1 demonstrates the minimum and maximum for each parameter, for the varying inlet profiles. Table 1.Comparisons between minimum and maximum of

each parameter. Inlet Profile Max.

Velocity (m/s)

Max. Wall Shear (Pa)

Min. Wall Shear (Pa)

Max. Pressure (Pa)

Min. Pressure (Pa)

Ramped 2.16E-08 1.03E-08

2.72E-11

1.76E-08 -1.35E-08

Sinusoidal 4.51E-09 3.01E-09

6.91E-12

2.18E-10 -1.59E-08

Constant (1µm/sec)

3.23E-05 1.39E-05

3.92E-08

4.03E-05 -5.29E-06

In terms of velocity, the smallest velocity magnitude was

found in the results of the sinusoidal profile. This is a result of the sinusoidal profile being more representative of natural breathing as compared to the ramped profile which represents the mechanical ventilation flow waveform. Hence, with this

sinusoidal flow representing a more natural breathing pattern there is less forced air, which creates less pressure and thus smaller resulting velocities. On the other hand, the largest velocity was found in the tracheal flow profile, 0.093 m/s. The minimum velocities were all 0 m/s which are expected as there will be regions throughout the model where no flow exists. The largest wall shear was found in the tracheal flow profile, whereas the smallest was found in the ramped flow profile.This is due to the fact that the flow waveform has spikes with sharp changes in the velocity throughout; hence as flow is simulated through the model over time, these sharp changes induce high wall shearing. The largest maximum pressure of 0.0227 Pa was found in the tracheal flow results. On the other hand, the smallest pressure was found within the ramped profile results.

4. DISCUSSION There has been a plethora of research on the lungs,

specifically on the alveoli, whether it has been particle deposition, stress distribution or fluid dynamics. Moreover Harding [11] performed a study on the flow in a terminal alveolar sac model using computational fluid dynamics and incorporating expanding walls. They focused on particle deposition and attaining an understanding of the fluid characteristics within the pulmonary region of the lungs. This gives an understanding of the flow in general breathing conditions. However for this study, the focus was based on mechanical ventilation and how these types of flow influence the fluid characteristics within the alveolar sac.

The results give us the magnitude of fluid characteristics with the highest velocities occurring in the turbulent model, and the largest pressures and wall shear occurring in the turbulent model, which is expected. However, excluding the tracheal flow and considering only the other ramped and sinusoidal profiles, the largest velocity, pressure and wall shear occurs within the ramped profile. On the other hand, the smallest pressure occurs in the ramped profile and the smallest wall shear occurs within the sinusoidal profile.

Figures 4 - 6 show the distribution of velocity, wall shear and pressure within the model, and from this study the regions where flow, pressure and wall shear are highest is where first signs of injury resulting from mechanical ventilation will occur. These parameters will be transferred from the fluid and onto the alveolar wall and will produce stresses/strains on the walls of alveoli.

As mentioned earlier with regards to the inlet velocity profiles, varying the flow profiles makes it possible to observe the differences in the resulting flow parameters. Further research will use the results from this study todetermine those stresses and strains that will be induced onto the walls. Hence, it will indicate when ventilation becomes dangerous, as the areas of high stresses/strains will be where tissue will undergo the most damage and cause inflammation. In addition a fluid/structure interaction will be performed in order to couple the fluid and solid and have load transfer take place simultaneously.


ACKNOWLEDGEMENT: The authors thank the National Science Foundation for supporting this work through grant CMMI-0969062.

REFERENCES [1] Florens, M., and Sapoval, B., and Filoche, M., 2011, “An Anatomical and Functional Model of the Human Tracheobronchial Tree”.J. Appl. Physiol., 110, pp. 756-763. [2] Ward, J.P.T, & Ward, J., & Leach, R.M. (2010). The Respiratory System at a Glance. (Rev. ed.), UK: Blackwell Publishing. [3] U.S. Department of Health and Human Services, Center for Disease Control and Prevention 2010: Summary Health Statistics for U.S. Adults: National Health Interview Survey, 2010. Washington, DC: U.S. Department of Health and Human Services, 2012. http://www.cdc.gov/nchs/data/series/sr_10/sr10_252.pdf (accessed Jan 26, 2012). [4]Kallet, R.H., Katz, J.A., 2003. “Respiratory system mechanics in acute respiratory distress syndrome.” Respir.Care.Clin. N. Am. 9(3), 297-319. [5] Cardenas Jr., Victor J., and Lynch, James E., 2006, “Mechanical Ventilation and Acute Respiratory Distress Syndrome” Semin Thoracic CardioVasc. Surg. 18, 8-12. [6]Pidaparti, R., and Koombua, K., 2011, “Tissue Strains Induced in Airways Due to Mechanical Ventilation,” Molecular & Cellular Biomechanics (MCB), 8(2), pp. 149-68. [7] De Ryk, Jessica, and Thiesse, Jacqueline, and Namati, Eman., and McLennan, Geoffrey., 2007, “Stress distribution in a three dimensional,geometric alveolar sac under normal andemphysematous conditions” International Journal of COPD. 2(1), 81-91. [8] Dailey, H.L., and Ghadiali, S.N., 2007, “Fluid-structure analysis of microparticle transport in deformable pulmonary alveoli”. J. Aersol Science. 38, 269-288. [9] FLUENT (ANSYS, Inc.). 2009. Fluent 12.1.4. [10] Przekop, Rafal, “Oxygen Transport in Human Alveolar Sacs” Warsaw University of Technology, Department of Chemical and Process Engineering. [11] Harding Jr., Edward M., and Robinson, Rida J., 2010, “Flow in a terminal alveolar sac model with expanding walls using computational fluid dynamics” Journal of Inhalation toxicology. 22(8), 669-678. [12] Haefeli-Bleuer, B., &Weibel, E. R. 1988. Morphometry of the human pulmonary acinus. The Anatomical Record, 220, 401–414. [13] Ochs, M., Nyengaard, J. R., Jung, A., Knudsen, L., Voig, M., Wahlers, T. et al. (2004).The number of alveoli in the human lung. American Journalof Respiratory Critical Care Medicine, 169, 120–124. [14] Li, Z., and Kleinstreuer. C., 2011, “Airflow analysis in the alveolar region using the lattice-Boltzmann method” Med Biol Eng Comput, 49:441–451.

[15] Dailey H.L., and Yalcin, H.C., Ghadiali, S.N., 2007, “Fluid structure modeling of flow-induced alveolar epithelial cell deformation” Computers and Structures, 85, 1066-1071. [16] Rausch, S.M.K., and Martin, C., and Borneamann, P.B., and Uhlig, S., and Wall, W.A. 2011, “Material Model of lung parenchyma based on living precision-cut lung slice testing” Journal of the Mechanical Behavior of Biomedical Materials. 4:583-592. [17] Rausch, S.M.K., and Haberthur, D., and Stampanoni, M., and Schittny, J.C., and Wall, W.A. 2011, “Local Strain Distribution in Real Three-Dimensional Alveolar Geometries” Annals of Biomedical Engineering. 39(11):2835-2843. [18] Harrington, L., and Prisk, G.M., and Darquenne, C., 2006 “Importance of Bifurcation Zone and branch orientation in simulated aerosol deposition in the alveolar zone of the human lung” Aerosol Science, 37, 37-62. [19] Polak, A.G., and Mroczka, J., 2004, “Nonlinear model for mechanical ventilation of human lungs” Computers in Biology and Medicine, 36, 41-58. [20] Clements, JA., and Hustead, R.F., and Johnson, R.P., et al. 1961. “Pulmonary surface tension and alveolar stability” J. App Physiol, 16, 444-450. [21] de Ryk, J., and Thiesse, J., and Namati, E., and McLennan, G., 2007. “Stress Distribution in a three dimensional geometric alveolar sac under normal and emphysematous conditions” International Journal of COPD. 2(1), 81-91.



November 2 – 3, 2012, Atlanta, Georgia USA

A REVIEW OF REHABILITATION STRATEGIES FOR STROKE RECOVERY

Daniel A Garcia, Rodrigo Arredondo, Melissa Morris, Sabri Tosunoglu Florida International University

Miami, Florida USA

ABSTRACT From neuronal to muscular disability, the field of

rehabilitation robotics has been working to find ways to increase the efficacy of treatment options that therapists provide on a daily basis. There are numerous systems of robot therapy that have been devised and studied in order to achieve this. It is the aim of this paper to focus on the rehabilitation strategies for upper limb motor control—post-stroke, the clinical effectiveness of said treatments along with data analysis methods, and summarize the road ahead in the field. Additionally, proposed methods not yet considered and tested will be discussed, including further integration of virtual reality.

INTRODUCTION Stroke, also known as cerebrovascular accident (CVA), is

the leading cause of long term disabilities and neurological damage in the United States and Europe [1]. A stroke causes permanent neurological damage or death. Annually, about fifteen million individuals worldwide suffer a stroke, killing about one third of them [2]. A stroke is caused by the loss of brain functions due to a disturbance of blood supply to the brain. This is usually caused by a blockage, a hemorrhage, or a lack of blood flow to the brain. As a result, the affected areas of the brain can no longer function properly, greatly affecting the individual. The result of a stroke may lead to an inability to move one or more limbs, restricted movement of one side of the body, and/or difficulty understanding and formulating speech [3], with approximately 25% stating that there was difficulty in reusing their arm sometime after the stroke [4]. Typically, the main consequence resulting from a stroke is the impairment of an arm and hand motor function.

As the population increases rapidly, so does the risk of a stroke or other neurologically damaging disease. Unfortunately, the availability of physical or rehabilitation

therapists does not increase as fast as the world’s aging population, which will need care because of the disabling conditions associated with the aging process [5]. Furthermore, one on one therapy sessions are expensive. For this reason, there is a need for alternative types of unattended, repetitive rehabilitation such as the implementation of robotic devices in therapy. This would ultimately increase the availability of treatments while creating more cost efficient ways to deal with the problem. Robotic devices can be used to provide safe and engaging therapy sessions for patients. A robot device can efficiently control the intensity of therapy while measuring changes of kinematics and forces.

Definition: To understand the history of the role robotics has played

as a treatment option for rehabilitation after stroke, it is important to define the different styles of assistance that a robot can provide. Assistive robotics (AR) refers to a general range of robots that are used for different locations including schools, hospitals and homes. The term socially interactive robotics (SIR) is used to refer to robots whose main tasks are destined to be interactions between operators and machines, and allow for the categorization of robots based on the type of interaction it has with the operator. Finally, the term socially assistive robotics (SAR) was coined to describe a robot whose goal encompassed many objectives including close interaction with its human operator, providing assistance or repetitive functions, and recording measurable data in rehabilitation and learning [6].

Expectations: As demonstrated in a paper released in 2003 by Van der

Loos et al. [5], for quite some time there have been great expectations for the development and use of rehabilitative robotics. Many of these expectations depend on a number of factors that influence how much impact and growth the rehabilitative robotic field will see in the future. One of the first requirements stated in the paper refers to the investment


of time and money by visionary developers and private sector investors respectively. Each of these will add valuable resources necessary for developing and advancing the field. Another important factor, as mentioned above, is the growing elderly population, which increases the incidence of stroke, thus necessitating aid and rehabilitation.

Acceptability: There have been many assessments as to the productivity

of using robot assisted rehabilitation in comparison to current modalities including direct physical therapy. Many of these assessments center on the fact that patients are able to relearn motor skills that are lost during a stroke by executing predetermined functional movements along with muscle control. Because the brain is able to adapt, these methods make use of neuroplasticity, the ability to adjust neuronal connections due to physical stimulants, and allows for the reestablishment of motor function [7]. While using the iPAM system, Jackson et al. determined that there was general positive feedback and acceptance of the device for future rehabilitative activity. This study concentrated on evaluating the user perception of the device and the results showed that all participants felt safe and secure while attached to the system, most were comfortable with its use, and most felt that the mobility in their arms had improved [7].

Challenges: While the motivation, expectations, and concerns over

acceptability are important for driving the field, the most critical factors are the challenges that must be overcome for general adoption by both user and therapist. The road for the implementation of robotics for use in modern therapy centers and in the home has several of these challenges that must be overcome in order to be feasible and functional. Research into mobilization, optimization, cost reduction, and weight reduction of current modalities has been extensive, but can still not be successfully implemented. Challenges must be identified in order to be addressed, and most of them have already been explored or are being researched. In order to make robotics an integral part of rehabilitation, the main challenges reside in the electromechanical implementation, neural control signals, the extraction of extent, and the clinical interactions [8]. Because the center of focus in this paper is for rehabilitative therapy, the main platforms observed are those that concentrate on the rehabilitation of neuronal muscle control, especially upper-limb movement.

As mentioned, overcoming the electromechanical implementation means that the robotic platform must be durable, easily maintained, light, flexible, and of course, have enough power to deliver comparable performance to normal muscle strength. The idea is to have a mechanism with the comparable size, weight, durability and strength of a normal limb. Because of the novelty of the field however, limited standards have been set for the application of robot-human interaction (HRI), and research into safety standards and redundant safety mechanisms necessary for robotics need to be further developed.

In terms of the neural control signals and the extraction of the person’s intent, the study of HRI is necessary, and must be

coupled with several studies of the human psyche and its perception of robot interaction. In this field, there are several topics that have been covered, from sensor usage to manual control mechanisms that are incorporated into the platform which allow for the direct or perceived control over the robot by its human operator. These also tie into the clinical aspect of how a person who is in charge of overseeing the patient’s progress can monitor, track, and interact with the robot. There is a need to allow for ease of use by both the operator and the care-giver.

DESIGN: While many of the designs in current robotic modalities

center on the many goals of providing repetitive exercise to the end user in the hopes of inducing the healing and relearning process, they approach the many challenges by concentrating on overcoming them individually. Two of the main design inputs that are focused on are the robot’s degrees of freedom (DOF) to simulate a natural arm, and the safety precautions to avoid bringing physical harm to the intended user. As will be demonstrated in the following examples, the emphasis of adding additional DOF and redundant safety checks becomes prevalent. But there is one other aspect that is introduced as well, virtual reality (VR) which gives a user an altered sense of HRI and has the benefit of tricking the mind into believing what it observes as well as stimulating the user with interactions that would otherwise be very boring and repetitive.

MIT-MANUS: The MIT-MANUS [9], shown in Figure 1, is a 2-DOF

robot manipulator developed by the Massachusetts Institute of Technology. 5-DOF can be added to the device by using an extension meant for wrist therapy. This device has been used to provide post-stroke therapy through a series of repetitive exercises. The design is meant to comply with a person’s movements while performing an exercise. Similar to a human therapist patient interaction, MIT-MANUS assists a patient to complete a movement. For this reason, there are safety concerns to be considered when designing the device. For instance, therapy sessions are to be administered and supervised by a human therapist. This ensures the device can be shut off in case of a mechanical malfunction occurring during a session. Second, the patients hand is attached to the device through a magnetic safety lock [10]. This enables the patient to pull free from device if need be with no external assistance. Last, the MIT-MANUS is designed to have low end-effector torques. A computer control system records the position, force, and velocity of the patient. This allows the computer to regulate the robot’s reactions to the patient ensuring gentle, compliant behavior.

This device has been used extensively in

neurorehabilitation research. Several improvements to this device have also been made in the past years. The MIT-MANUS has shown a decrease in upper-limb impairment of patients after being discharged.


Figure 1. MIT-MANUS in a clinical trial with VR [9]

ARMin: The ARMin [11], shown in Figure 2, is a 4-DOF arm

rehabilitation robot with a few extra freedoms located on the wrist area that was developed to assist therapists in delivering rehabilitation with consistent, repetitive exercise. The design was based around maintaining the user’s normal range of motion in relation to the activities of daily life (ADL). Several safety considerations were taken into account when designing the robot as well. The first area of concern was at the physical level; the robot would not contain any sharp edges in its construction, and no joint would extend beyond the anatomical range of motion for a human limb. The second level of safety was implemented as an active detection method. Several position sensors would detect the malfunction of a controller. The final level of security with the design itself was within the software, where routines were implemented that would monitor for abnormal events and cutoff power immediately in the case of detection. Along with all of these safety factors, four emergency stop buttons were added to the device, one of which was a dead-man’s switch which was given to the therapist overseeing the session.

Although the device in the study underwent a limited trial of short sessions, the improvement of the patient’s motor functions could be observed through the decrease in support and aid that the robot had to give the patients. There was also a noticeable increase in the range of motion of the upper limb as well as increased joint coordination [11].

CADEN-7: Referenced in the name, the CADEN-7 [12] is a 7-DOF

robot that stands for “cable-actuated dexterous exoskeleton for neuro-rehabilitation” and is a stem from earlier models that had 1 and 3 DOF. It is shown in Figure 3. The initial designs were used as a probe study to show proof of concept of control using surface electromyography (sEMG), which was successful, thus leading to the design and construction of the CADEN-7. The CADEN-7 was unique from others in the field at the time because of its incorporation of the control method, which allowed for detection of electrical signals on the surface of a user’s skin, that are generated by the action

potentials that are fired in the muscular cells when initiating movement.

This method of control even has the ability to predict a user’s movement before initiation, and thus synchronize the robot to act almost seamlessly with a user’s arm. Aside from the design that implemented the 7-DOF to match a study of ADL, there was also a focus on safety which resembled that of the ARMin. The CADEN-7 also employs a three level safety protocol, starting at the mechanical level, and then moving on to the electrical and software levels. The focus was on avoiding hyperextension and erratic movement, as well as emergency stop buttons within reach of both a user’s hands and feet. Other considerations that the authors stated were that until higher efficiency power-to-weight ratio motors and structural materials are developed, wearable robotics will remain fixed to some form of external structure.

Figure 2. ARMin in a clinical setting [11]

L-EXOS: The L-EXOS [13] system, shown in Figure 4, truly differs

from the rest in its consideration and incorporation of VR. The main focus of the L-EXOS has been in the increase in repeatability of exercise execution, real time measurement and visualization of physiological data, and immediate assessment of rehabilitation progress from stored data. The main design inputs focused on an exercisable force at the end-effector of 100N in every direction, redundant sensors for patient safety,


modular design for maintenance ease, limited back-drivability, configuration change from left to right arm, 5 DOF with four being actuated, position sensor at every DOF, and isomorphism with human arm kinetics. These considerations show the progress in exoskeleton design and extend into providing actual values for safety. Of the important values, the maximum output velocity should not exceed 10rpm and the maximum output torque should not be over 120Nm. Other safety implementations included joint torque sensing and redundant checks of system integrity.

Figure 3. CADEN-7 performance display [12]

After a limited study was completed with the finished

prototype, they found that there was no general correlation with the reported benefits and the performance improvements. However, it was observed that the upper limbs can be positively stimulated by exoskeleton devices like the L-EXOS, even if there is a limitation due to the number of DOF [14].

MIME: The MIME is a 6-DOF industrial robot manipulator used

in rehabilitation of upper-limb motor control [15]. The main goal of the MIME device is to improve the range extent of the paretic limb through a series of exercises. This device is able to apply forces to a paretic limb as motor control is recovered using a modified fore arm splint. This allows the manipulation of the forearms through a large range of motion in a three dimensional space. Since the forearm is restrained, safety measures had to be considered in the design in order to prevent injury to limbs. For instance, the MIME is able to perform naturalistic arm postures during reaching exercises. Second, the MIME has four program modes used to accommodate the abilities of each patient [16]. These program modes are passive, assistive, resistive, and bilateral. Bilateral mode consists of repetitive movements of the paretic limb to the mirror image position of the unimpaired limb. Lastly, the device is also able to measure the forces and torques between

the robot and patient throughout each task. This is necessary to provide exercises in a range of forces that can be handled by the patient. Ultimately, this feature reduces the risk of further damage to the paretic limb by excessive forces.

The MIME, shown in Figure 5, is one of the many devices currently in use for upper-limb rehabilitation. Several techniques and exercises have been tested using this device in hopes of improving upper-limb reach extents of the shoulder and elbow. To this date, MIME has shown significant improvement of the upper-limb reach extents in post-stroke patients.

Figure 4. L-EXOS with the utilization of VR [13]

CLINICAL: Several studies have shown that there is an improvement

in upper-limb motor control using robot assisted therapy. These studies show that patients suffering from mild to severe impairment who received robotic therapy had improvements in muscle strength and motor coordination of the affected limb. To more effectively measure and understand the progression of each individual patient, several scoring systems have been developed. Their purpose is to assess the severity of motor control and dependency of post-stroke patients. Some of these systems are the Barthel Index (BI) test, Wolf Motor Function Test (WMFT), and the Fugl-Meyer Assessment (FMA) which are explained in section 3.2. In the following sections, different devices and techniques are also reviewed in order to demonstrate the different types of therapy available. Mainly focusing on wrist, elbow, and shoulder therapy, these therapies consist of active-assistive exercises, planar reaching tasks, and progressive-resistive training. Virtual and auditory techniques are also reviewed in order to explore their potential in robot aided therapy.


Figure 5. An example of the MIME system [18]

TREATMENT: Once the loss of motor control has been assessed, the

therapist can then look at the type of treatment best suited for the patient. Providing the right treatment could ultimately improve the recovery of motor control in the long term. In recent years, robot assisted therapy has become available as an alternative to traditional therapy. Robot therapy mainly focuses on performing repetitive tasks over a period of time which helps patients regain control, strength, and range of motion of the affected area. Although a broad area, robot therapy can be classified into three types of treatments: passive therapy, assisted therapy, and resisted therapy [17]. The use of these treatments depends on the performance of the patient receiving therapy.

Passive Therapy is administered by the therapist while the patient is relaxed. This approach is used to assess the range of motion and flexibility of the limb being treated. Mainly used in upper limb extremities, this treatment consists mainly of stretching and contracting exercises. This exercises help assess the range of motion a patient has prior to being exposed to more challenging tasks. It has also been shown that this type of treatment is an essential component in reducing muscle stiffness in spastic patients [18]. The implementation of a robot exoskeleton while performing these tasks provides valuable information to the therapist. As shown in a recent

study [19], robot assisted treatments can provide a wide variety of muscle activity data as a task is being performed. This can be useful to the therapist by providing the information necessary to treat a specific disorder and plan a path for treatment.

Assisted Therapy is used when a patient is not able to perform an exercise. As a patient struggles completing a task or a specific movement, external forces are applied. These forces can be applied by a therapist or by a robotic device. Ultimately, patients should be able to complete a task or improve their mobility and range of motion. For instance, in a recent study [20], this technique was used to assist patients during shoulder and elbow exercises. If a patient was not able to move to a desired target, the robot would provide assistance in order to reach the target. If a patient could perform the majority of the task but could not complete it, the robot would intervene. The robot would then attempt to read the patients intended movements and give assistance as needed. In this particular study, patients subjected to robot assisted therapy saw an improvement on the Fugl-Meyer Assessment scale after treatment was completed.

Resistive Therapy involves a patient performing tasks while one or more opposing forces are being applied. The opposing forces can be applied by the therapist or by the robot mechanism. These forces can be in the form of gravity, constraints within the environments, or manipulation of the robot or therapist. In the study [20] mentioned earlier, subjects were also administered resistive therapy. When a session had been completed, the amount of force used is recorded. The next session starts with a small increment of the last recorded force. This is technique is intended to improve overall, long-term strength of the treated arm.

In order to effectively evaluate the improvement of a patient, assessments should be given at constant time intervals. Tracking improvement through therapy is essential in order to determine whether the current type of treatment is effective or whether it needs to be changed. This is important to prevent treatments that could potentially decrease the recovery rate of a patient. Similarly, if there is a consistent and significant improvement, a patient could potentially be introduced to more rigorous exercises. Studies [20][21][22] have also shown that implementing vision and auditory feedback can be an essential tool in therapy. In an auditory feedback experiment [21] enhancing the performance of the patient was achieved by this method. Patients were subjected to trials with and without auditory feedback. The results showed that patients receiving auditory feedback had better results than in trials with no auditory feedback. Similarly, there is research [22] implementing visual aid assistance to patients during treatment that yielded positive results. Tasks were performed with and without visual feedback on the same group of individuals. Results showed an improvement in task performance by the patients who received visual feedback. Combining visual and auditory feedback to current robot devices could benefit the recovery of a patient. These are some of the tools that can be applied in robot assisted therapy to further improve the efficacy of treatment. In the next section, scoring systems are discussed. These scoring systems are used to further research


in the area of robotics and rehabilitation as well as the recovery rate of patients as they are exposed to different types of treatments.

SCORING/ANALYSIS: In order to better assist a post-stroke patient, the severity

of the damage the individual sustained as a result of stroke must be assessed. Properly identifying the areas affected by and the amount of damage from a stroke is important in providing the right type of therapy. For instance, once the affected areas have been determined, therapy sessions can be focused on those areas. The amount of training needed can also be reflected by the score of a patient. Typically, a therapy session lasts for an hour a day. Therapy can be given for one or several days a week. Usually, treatment lasts for an average of three months. The following scoring systems are used in order to determine the severity and type of therapy best suited for a patient.

The Barthel Index (BI) evaluates the functional independence of a patient [23]. This assessment is mainly used to measure the amount of assistance a patient needs while performing ten activities of daily life. These tasks are related to the self-care and mobility of the individual. The assessment is rated by observation of the person being given the therapy or the one providing it. The lowest score possible is a zero, which indicates a patient is totally dependent. Such an extreme case may suggest an individual cannot perform the simplest of tasks without some type of assistance. The highest score is 100 indicating a patient is able to perform all tasks given without any type of assistance. The tasks performed by the patient are feeding, grooming, bathing, dressing, bowel movement, bladder control, toilet use, object transfer, limb-mobility, and going up and down a set of stairs. It has been shown that the earlier a patient shows improvement in the Barthel Index, the better the outcome is, after 6 months after the stroke [24]. The BI can also be applied to different types of injuries involving loss of motor control in their upper and lower extremities.

The Wolf Motor Function Test (WMFT) was first developed to evaluate changes in upper-limb impairment in patients after a stroke [25]. The WMFT is also used to quantify motor function not only after a stroke, but also after other types of brain injuries. This test is done by timing a patient while performing a total of 15 tasks. These tasks are divided into two categories from simple movements to more difficult tasks. Tasks 1 to 6 involve timed joint-segment movements [26]. The tasks used to evaluate the patient consist of moving the forearm to a table, moving the forearm towards a box, extending the elbow, moving a hand to a table, and lastly, moving a hand to a box movements. The remaining tasks consist of timed integrative functional movements [26]. The tasks for this part of the assessment involve reaching and retrieving, lifting a can, lifting a pencil, picking up a paper clip, stacking checkers, flipping cards, turning a key in lock, folding a towel, and lifting a basket. Each task is given a time limit of 120 seconds in order to complete. A higher score is indicative of worse motor control of the joints and extremities targeted by each task. The WMFT scores are commonly used

because it gives important insight into the level of function and potential motor recovery of a patient [27].

The Fugl-Meyer Assessment (FMA) assessment consists of several tests used to measure the impairment of a patient. The assessment also addresses the amount of pain, sensation, range of motion and the balance of a patient as a task is performed. In order to assess balance, the patient performs sitting and standing tasks and is scored accordingly. Sensation is evaluated by touching the extremities of the patient. Range of motion is tested on eight joints, four in each extremity through series of exercises. As each joint moves through its available range of motion, the joint pain is assessed and scored accordingly. For the upper extremity part of the assessment, the test is focused on the shoulders, elbows, forearms, wrists, and hands. For lower extremities, the hips, knees, and ankles are the main focus [28]. Each task receives a score ranging from 0 to 2. A 0 indicates the task was not completed, a 1 the task was partially completed, and a 2 for a task performed completely. Fugl-Meyer assessment is one of the most common tests used in order to assess recovery of physical strength after a stroke. This is because unlike the BI, the Fugl-Meyer assessment can pinpoint the affected areas in a patient in order to provide adequate treatment. The Fugl-Meyer Assessment Score sheet can be found in the Appendix section.

These scoring systems and similar types of assessments can also be applied to evaluate the neurological damage of a patient after suffering from a spinal cord injuries, brain injuries, or similar injuries. The scoring systems described above are among the most commonly used to evaluate the impairment of a post-stroke patient. For stroke related impairments, it is recommended to administer the evaluation within the first three months of the episode. It has been suggested that there is a process of spontaneous recovery within the first month post-stroke and then gradually diminishes over six months [24]. Early assessment of a patient allows the therapist to initiate therapy during the first month, providing proper treatment to the affected area.

Several studies implementing the use of the mirror image movement enabler (MIME) have shown the positive aspects of robot assisted therapy. In two recent studies [29][30], two groups were exposed to different rehabilitation techniques. This was done in order to further investigate robot assisted therapy versus traditional therapy six months after the stroke. The treatment takes place over a two month period with a total of twenty-four one hour sessions. All patients receive a FMA score in order to evaluate motor control recovery before the therapy program starts, after discharge, and at a six month follow up. One group is exposed to traditional human therapist patient interactions and is referred to as the control group. The second group is exposed to robot assisted therapy. The exercises provided to the patients can be performed by therapist and robot alike. The main difference between the exercises is the type of movements, modes of assistance, and number of repetitions in a given session. Test results showed a quick improvement of the impaired upper extremities on patients receiving robot assisted therapy. The rate of recovery in respect to strength, reach, and overall mobility of the


affected limbs is seen to improve faster in the robot assisted group compared to the control group. As the therapy progressed toward the second month, no significant difference between the recovery of the control group and the robot assisted group could be seen. At the six month follow-up, the robot assisted group clearly showed a greater gain in upper-body control than the control group [29]. Recovery of strength and motor control improvement after six months was maintained by the robot assisted group. Although both groups saw a significant improvement in upper limb function, this and other studies suggest that robot assisted therapy is potentially more cost effective and beneficial for the patient. This pattern can be seen in several studies that further support the use of robot assisted therapy. However, more control is needed as there were many discrepancies within the studies which may have influenced the usability of the data, including pre-conditioned therapy patients, age difference and attitude, and length of therapy sessions.

DISCUSSION: As mentioned in previous sections, the field of

rehabilitative robotics has come a long way from its early days—and even from the MIT-MANUS. There are some limitations due to the current technology available, such the size and weight of actuators, as well as materials in terms of their strength to weight ratio. There are still many areas that can be focused and improved on, however, and optimizations are never complete. Systems can always be made more efficient. One of the areas that has seen much attention is the study of ADL and what the appropriate setup and DOF are required to meet the rigorous and free movement of a normal arm. The most complex system explained here is a 7-DOF robot, but many exist with even higher DOF. With more DOF, the complexity of the device increases, and so does the need to drive each joint, which adds to weight and bulk. Further analysis is needed to find the optimal setup and structure that can be comparable to ADL. The other area that the aforementioned robots and many others focus heavily on is safety. Although there are several precautions already implemented, and multi-layered redundancy checks already in place, no standards exist for value tolerances such as maximum safe speed, torque, and extension. The L-EXOS, however, makes an attempt at setting some parameters to adhere to.

There are also ideas for further integration of sensor detection coupled with robot response and virtual reality. As it stands, most of the apparatuses that utilize VR do it in a manner of a “video game” style interface, where the user sits strapped into the robot watching some form of image or moving image on a screen. The image either initiates a response or accepts an input and generates a response in accordance to the input. VR has proven to be effective in stimulating patients to be more enthusiastic about the treatment, as the environment lends itself to creating a challenge to overcome. But, as with everything, improvements can always be made, and one idea to stimulate more interest is to completely submerge the subject into VR where all sight and sound is separated from the current surroundings and replaced with a virtual environment. The idea is to isolate

sight and sound with a headpiece that displays an image to the user with a first person perspective, where the actions of the virtual character represent the movement initiated by the user of the robot. With assistance provided by the robot, the user can attempt virtual tasks with the impression that they are accomplished without the visual or audible presence of the actual robotic device. Because the brain can be affected by physical and sensory stimulants thanks to its neuroplasticity, using this method may recondition the brain at an increased rate, relying on the concept of self-sufficiency and encouraging self-confidence.

CONCLUSION Over the years, many techniques and devices have been

developed to regain motor control of upper and lower extremities after a stroke. Although a broad area, robots have increasingly been used for research and testing new hypothesis in the area of rehabilitation. As mentioned earlier, there is a window of exponential recovery patients have to take advantage off. It has been suggested that this time of recovery takes place during the first three months of an episode and gradually diminishes six months after the stroke. At this time, strength, range of motion and mobility can be recovered to a great extent. Therefore it is essential that during this time, patients exercise regularly in order to regain motor control. As shown by several studies, robot assisted rehabilitation is an effective way of treatment which would ultimately allow a patient to regain a large portion of their motor control. Implementing robot assisted exercises with regular therapy has also been shown to improve the overall independence of post-stroke patients. As more individuals suffer from strokes and the population grows rapidly, robot devices can supply the demand for more efficient methods of treatment. The cost effective aspect could also allow treatment for individuals that do not have the resources necessary for a therapist. Even though robot assisted therapy results are promising, further research needs to be conducted. Increasing motor control of a patient using robot devices is relatively new, and more techniques need to be tested. What can be agreed on, is that robot assisted therapy could potentially become the best type of treatment not only for stroke victims but for a wide variety of neurological injuries.

REFERENCES [1] Feigin, VL. Stroke Epidemiology in the Developing World. Lancet Journal 2005. [2] Rosamond, W., Flegal, K., Friday, G., Furie, K., Greenlund, K., and Haase, N. Heart Disease and Stroke Statistics. American Heart Association Statistics Committee and Stroke Statistics Subcommittee. 2007. [3] Donnan, GA, Fisher M, Macleod M., and Davis, S. M. Stroke. Lancet Journal 2008. [4] Geddes, J. M., Fear, J., Tennant A., Pickering A., Hillman M., Chamberlain M. A. Prevalence of Self-Reported Stroke in the Population of Northern England. Journal of Epidemiol Community Health Med 1996. [5] Machiel Van der Loos H. F., Mahoney R., Ammi C. Great Expectations for Rehabilitation Mechatronics in the Coming


Decade. 8th International Conference on Rehabilitation Robotics 2003. [6] Feil-Seifer D., and Mataric, M. J. Defining Socially Assistive Robotics. 9th International Conference on Rehabilitation Robotics 2005 [7] Jackson, A. E., Makower S. G., Culmer, P. R., Holt, R. J., Cozens, J. A., Levesley, M. C., and Bhakta, B. B. Acceptability of Robot Assisted Active Arm Exercise as Part of Rehabilitation After Stroke. IEEE International Conference on Rehabilitation Robotics 2009. [8] Dellon, B., and Matsuoka, Y. Prosthetics, Exoskeletons, and Rehabilitation [Grand Challenges of Robotics]. IEEE Robotics & Automation Magazine 2007. [9] Hogan, N., Krebs, H. I., Charnnarong, J., Srikrishna, P., and Sharon, A. MIT-MANUS: A Workstation for Manual Therapy Training. Newman Laboratory for Biomechanics and Human Rehabilitation MIT 1992. [10] Krebs, H. I., Dipietro, L., Levy-Tzedek, S., Fasoli, S. E., Rykman-Berland, A., Zipse, J., Fawcett, J. A., Stein, J., Poizner, H., Lo, A. C., Volpe, B. T., and Hogan, N. A Paradigm Shift for Rehabilitation Robotics. IEEE Engineering in Medicine and Biology Magazine 2010. [11] Nef, T., Mihelj, M., Colombo, G., and Riener, R. ARMin - Robot for Rehabilitation of the Upper Extremities. IEEE 6th International Conference on Rehabilitation Robotics 2006. [12] Perry, J. C., Rosen, J., and Burns S. Upper-Limb Powered Exoskeleton Design. IEEE/ASME Transactions on Mechatronics 2007. [13] Vertechy, R., Frisoli, A., Dettori, A., Solazzi, M., and Bergamasco, M. Development of a New Exoskeleton for Upper Limb Rehabilitation. IEEE International Conference on Rehabilitation Robotics 2009. [14] Frisoli A., Bergamasco, M., Carboncini, M. C., Rossi, B., Greco, G., Montagner A., and Procopio, C. Robotic Assisted Rehabilitation in Virtual Reality with the L-EXOS. Studies in Health Technology and Informatics 2009. [15] Kahn, L. E., Lum, P. S., Rymer W. Z., and Reinkensmeyer, D. J. Robot-Assisted Movement Training for the Stroke-Impaired Arm: Does it Matter what the Robot Does? Journal of Rehabilitation Research & Development [16] Hesse, S., Schmidt, H., Werner, C., and Bardeleben, A. Upper and Lower Extremity Robotic Devices for Rehabilitation and for Studying Motor Control. Current Opinion in Neurology 2003. [17] Lum, P., Reinkensmeyer, D., Mahoney, R., Rymer, W. Z., and Burgar C. Robotic Devices for Movement Therapy After Stroke: Current Status and Challenges to Clinical Acceptance. Top Stroke Rehab 2002. [18] Schmit, B. D., Dewald, Julius P. A., and Rymer W. Z. Stretch Reflex Adaptation in Elbow Flexors During repeated Passive Movements in Unilateral Brain-Injured Patients. Archives of Physical Medicine and Rehabilitation 2000. [19] Ueda, J., Ming, D., Krishnamoorthy, V., Shinohara, M., and Ogasawara, T. Individual Muscle Control Using an

Exoskeleton Robot for Therapy Muscle Function Testing. IEEE TNSRE: UEDA 2010. [20] Fasoli, S. E., Krebs, H. I., Stein, J., Frontera, W. R., Hughes, R., and Hogan N. Robotic Therapy for Chronic Motor Impairments after Stroke: Follow-Up Results. Archives of Physical Medicine and Rehabilitation 2004. [21] Rosati, G., Oscari, F., Reinkensmeyer, D. J., Secoli, R., Avanzini, F., Spagnol, S., and Masiero, S. Improving Robotics for Neurorehabilitation: Enhancing engagement, Performance, and Learning with Auditory Feedback. IEEE International Conference on Rehabilitation Robotics 2011. [22] Vergaro, E., Casadio, M., Squeri, V., Giannoni, P., Morasso, P., and Sanguineti, V. Self-Adaptive Robot Training of Stroke Survivors for Continuous Tracking Movements. Journal of NeuroEngineering and Rehabilitation 2010. [23] Loewen, S C., and Anderson, B. A. Reliability of the Modified Assessment Scale and the Barthel Index. Journal of the American Physical Therapy Association 1988. [24] Krakauer, J. W. Motor Learning: Its Relevance to Stroke Recovery and Neurorehabilitation. Current Opinion in Neurology 2006. [25] Wolf, S. L., McJunkin, J. P., Swanson, M. L., and Weiss, P. S. Pilot Normative Database for the Wolf Motor Function Test. American Congress of Rehabilitation Medicine and the American Academy of Physical Medicine and Rehabilitation 2006. [26] Lo, A. C., Guarino, P. D., Richards, L. G., and Haselkorn, J. K., Wittenberg, G. F., Federman, D. G., Ringer, R. J., Wagner, T. H., Krebs, H. I., Volpe, B. T., Bever, C. T. Jr., Bravata, D. M., Duncan, P. W., Corn, B. H., Maffucci, A. D., Nadeau, S. E., Conroy, S. S., Powell, J. M., Huang, G. D., and Peduzzi, P. Robot Assisted Therapy for Long-Term Upper-Limb Impairment After Stroke. New England Journal of Medicine 2010. [27] Wolf, S. L., Catlin, P. A., Ellis, M., Archer, A. L., Morgan, B., and Piacentino, A. Assessing Wolf Motor Function Test as Outcome Measure for Research in Patients After Stroke. American Heart Association, Inc 2001. [28] Sanford, J., Moreland, J., Swanson, L. R., Stratford, P. W., and Gowl, C. Reliability of the Fugl-Meyer Assessment for Testing Motor Performance in Patients Following Stroke. Journal of the American Physical Therapy Association 1993. [29] Lum, P. S., Burgar, C. G., Shor, P. C., Majmundar, M., and Van der Loos, M. Robot-Assisted Movement Training Compared with Conventional Therapy Techniques for the Rehabilitation of Upper-Limb Motor Function After Stroke. Archives of Physical Medicine and Rehabilitation 2002. [30] Burgar, C. G., Lum, P. S., Scremin, A. M. E., Garber, S. L., Van der Loos, M., Kenney, D., and Shor, P. Robot-Assisted Upper-Limb therapy in Acute Rehabilitation Setting Following Stroke: Department of Veterans Affairs Multisite Clinical Trial”. Journal of Rehabilitation Research & Development 2011.




THE EFFECT OF STENT DESIGN ON HEMODYNAMIC FLOW IN PATIENT-SPECIFIC DISEASED CAROTID BIFURCATION ARTERIES

Rodward L. Hewlin, Jr. and John P. Kizito

North Carolina A&T State University Department of Mechanical Engineering

1601 East Market Street Greensboro, NC, USA

Email(s): [email protected] and [email protected] Phone: (336) 334-7620 ext. 315

ABSTRACT The present work examines the role of stent design on

hemodynamic flow in patient-specific diseased carotid bifurcation arteries for carotid artery stenting (CAS) procedures. Vascular stenting procedures have been accepted as a standard form of treatment of cardiovascular disease, the most common form being atherosclerosis. Despite the widespread use of stents in treating atherosclerosis, these medical devices are not without their problems. In the present work, three in-house stent designs (Stent designs: A, B, and C) are evaluated. The in-house stent designs are evaluated in simplified arteries and patient-specific arteries. The simplified arteries are straight constant diameter tubes and the patient-specific arterial geometries are obtained from reconstruction of CT scan images of diseased patients. Hemodynamic results (wall shear stress (WSS), radial stresses, and vorticity magnitudes) were compared for simplified and patient-specific stented arteries to determine if simplified stented geometries can predict the flow characteristics and arterial wall forces experienced in patient-specific stented carotid bifurcation arteries. Substantial differences in hemodynamic parameters were found to exist which confirms the effect of stented core geometry, and stent design, specifically strut positioning and bridge pattern design on hemodynamics.

keyword(s): atherosclerosis, carotid bifurcation artery, patient-specific arteries, radial stress, simplified arteries, strut position, vorticity, and wall shear stress.

INTRODUCTION Cardiovascular disease (CVD) remains to be the leading

cause of morbidity and mortality across the world and improved methods for cardiovascular disease management are deeply needed [1]. The most common form of cardiovascular disease is atherosclerosis. The carotid bifurcation artery, where the common carotid artery (CCA) branches into the internal carotid artery (ICA) and external carotid artery (ECA) is a common site of atherosclerotic disease [2]. Stenosis has long been known to be related to the incidence of ischemic stroke and ischemic transient attack. Carotid endarterectomy (CEA) was considered the “gold standard” for severe carotid artery (CA) stenosis [3], but recently carotid artery stenting (CAS) is emerging as a safe and cost-effective alternative to CEA [4]. The treatment procedure is nominally invasive and clinical studies indicate superior long term patency of stent usage.Although stents are widely used in treating atherosclerosis, these medical devices are not without their problems.

Despite the widespread use and numerous clinical studies, stents are implanted into arteries without a priori understanding of the associated fluid dynamics [5]. Although computational tools such as finite element analysis (FEA) are used to investigate several aspects of vascular stenting, such as the evaluation of interventional techniques [6], impact of plaque composition on vessel wall shear stress (WSS) [7], impact of stent expansion on vessel wall [8], few FEA studies are available on CAS [9-10]. Furthermore, few CFD studies are available on evaluating the effect of stent design (strut positioning and bridge pattern) on hemodynamic flow for CAS procedures [11]. However, a number of CFD studies have been


conducted in the effort of gaining an understanding of the flow disturbance experienced by the presence of stents in arterial vessels [12-16].

An experimental and two-dimensional computational flow analysis using custom stent models was performed to examine the effect of wire spacing and diameter on stagnation zones were studied by Santamarina et al., [17]. The authors reported that stent geometry had a substantial effect on arterial hemodynamics. Steady-state three-dimensional CFD simulations were performed in a Palmaz-Schatz slotted-tube stent by Faik et al. [23]. The experiment was done using data from in vivo measurements of canine left anterior descending (LAD) coronary artery diameter and blood flow velocity. The authors reported that regions of low WSS are localized around the stent struts. They also reported that reducing the number of struts and strut thickness resulted in a reduced percentage of arterial wall area exposed to low WSS. The opposite was observed if strut width was decreased.

Pulsatile and non-Newtonian blood flow through a stent with a helical strut matrix was studied by Banerjee et al.[18]. The developed model was used to evaluate the effect of entrance flow where the stent is placed at the entrance region of a branched coronary artery stent geometry. Recirculation zones were identified immediately upstream and downstream of each strut intersection. In the study of Gervaso et al. [19], the expansion of four different stent designs were modeled against plaque in the artery using FEA and used the expanded geometry to evaluate hemodynamics. Stent models were expanded against a simplified plaque, with a solid mechanics analysis, and then subjected to a fluid flow simulation under pulsatile physiological conditions. Spatial and temporal distribution of arterial wall shear stress (WSS) was investigated after the expansion of stents of different designs and different strut thicknesses.

Several clinical studies have highlighted that the success of CAS is strongly dependent on the operator capability and should be reinforced by proper selection of patients and devices [8, 20]. Moreover, clinical data suggests that CAS outcomes are essentially related to anatomic considerations [21], specifically patient-specific data. Stent design issues associated with CAS procedures include: (1) adequate rigidity to resist the compressive forces from the vessel wall and dislodging/migration forces from blood flow (from WSS); (2) arterial vessel scaffolding properties: stents must be capable to hold open the vessel and scaffold material (plaque) against the vessel wall; (3) minimal longitudinal contraction when expanded; (4) minimal shearing between the stent and tissue during expansion because this exposes the vessel of its endothelial cell lining; and (5) significant alterations in hemodynamics leading to particular zones which could be susceptible to smooth cell proliferation and restenosis [22-24]. All of these considerations are challenging to meet in any one stent design. As a result, there is a significant need to understand the effect of stent design characteristics (strut positioning and bridge pattern) on vessel hemodynamics.

In the present work, simplified stented core hemodynamic results are compared to patient-specific hemodynamic results to determine whether CFD simulated simplified stented core geometries can predict the disturbed flow characteristics and arterial wall forces experienced in stented arteries vs. the stented patient-specific carotid artery image-based CFD approach. The following issues are addressed: (1) the influence of stent strut positioning and bridge pattern design on WSS, and (2) the influence of stent strut positioning and bridge pattern design on the normal (radial) forces (cyclic/fatigue loading), and the magnitude of vorticity (de-mixing) that occurs in the stented arterial fluid cores. The main limitations found in earlier artery stenting simulation literature are: (1) assuming that stents and the WSS and normal forces acting on the stents are axisymmetric; (2) neglecting the varying cross-sectional area of the arterial wall upstream and downstream of the stented region; and (3) simplifying the arterial fluid core to cylindrical models.

MATERIALS AND METHODS The three in-house stent design patterns evaluated in the

present work are similar to stents manufactured by Abbott Laboratories (Abbott Park, Illinois, U.S.A.) and Biosensors (Europe, SA (BESA), Switzerland). The geometric details of each stent are provided in Table 1. To make a design comparison, representative geometries for each stent are constructed with the same strut: diameter (5 mm), length (24 mm), and thickness (0.24 mm). The strut spacing for each stent design evaluated is representative and differs to some extent from commercially available stent designs. Figure 1 shows the flattened out geometries for the three in-house stent designs. The flattened out geometries shown in Figure 1 were created in Solidworks. The flattened geometry was sketched on a plane, extruded to the required thickness, and then converted to a circular meshed tube (stent) using a wrap feature. For CFD simulations in the present work, the flow in the carotid bifurcation artery is Newtonian, laminar, and incompressible. The assumptions used in the present work have also been used in the work of [17, 25].

Table 1: Geometric details of stent models. Model label A B C Number of Columns 10 10 10 Number of Struts 42 10 72 Outer diameter 5 mm 5 mm 5 mm Strut thickness 0.24 mm 0.24 mm 0.24 mm Length 20.25 mm 20.25 mm 20.25 mm Aspect Ratio (e/d) 0.048 0.048 0.048 Aspect Ratio (e/L) 0.012 0.012 0.012 Aspect Ratio (d/L) 0.250 0.250 0.250

Blood is modeled by the incompressible Navier-Stokes

equations with blood density specified as 1060 kg/m3 and the corresponding dynamic viscosity 0.0032 Pa s⋅ . The commercial software ANSYS was used for geometric meshing and FLUENTTM was used to solve the Navier-Stokes equations


in the finite volume formulation. Numerical simulations were performed for simplified arterial stented cores and patient-specific arterial stented cores. Figure 2 shows the simplified arterial stented cores and patient-specific arterial cores that were evaluated in the present work. The simplified arterial stented cores consisted of a half-plane stented segment since the stent is axisymmetric as shown in Figure 2b. The patient-specific stented core consisted of an arterial geometry obtained from a computed angiographic (CT) scan of a 50 year old male as shown in Figure 2c. The image contained a point cloud, which defines the surface geometry of the carotid bifurcation by discrete domains. The image was processed in Solidworks CAD software using a curve creation feature similar to the work of R.L. Hewlin, Jr. and J.P. Kizito 2011 [25].

(a)

(b)

(c)

Figure 1: Flat geometries (to the left) and stent geometries (to the right) for stent designs under

investigation: (a) Stent Design A, (b) Stent Design B, and (c) Stent Design C.

The curve creation feature united the point cloud with 3-D

surface splines. This technique was repeated at each domain until the entire arterial geometry was defined. Each 3-D spline domain was then united using a lofting and boundary boss feature until the final solid product was obtained. The arterial wall at the stented region for both the simplified arterial stented core and the patient-specific arterial stented core is straight with constant diameter.The three dimensional stented regions were created in Solidworks with a solid cylinder dimensioned 1.5 times the length and the diameter of the core. The core

geometry represents full strut exposure to the blood flow. The stent geometry and the solid cylinder were united using the mating feature then removed using a Boolean operator in which the geometry of the stent is removed from the cylinder and the stent geometry impression remains in the cylinder. The models are discretized and used as computational domains. Grid meshing levels were performed at three levels: coarse (132,226), coarse (155,998) and medium (306,085).

A sinusoidal velocity waveform boundary condition was specified at the inlet and the outlet(s) was modeled as pressure outlets at 12 kPa (90 mmHg). The No-slip boundary conditions were imposed at the walls. The flow velocity amplitude in the inlet of each stented core was set to 0.2 m/s and a heart rate of 168 beats/min (2.8 Hz) to simulate normal cardiac behavior.The comparison of the velocity magnitude at the ICA for arterial stented core geometric models resulted in a 0.32%

Figure 2: Arterial stented fluid cores: (a) simplified stented

core cross-sectional area, (b) simplified stented arterial core, and (c) patient-specific stented arterial core.

difference in velocity in a grid independence study. With such a small difference, simulation results were independent of the computational mesh when the disparity between mesh spacing varying densities was less than 1%. For numerical solutions, the pressure coupling solving technique was implemented. The WSS, normal stresses, and vorticityvalues are calculated on fluid domain surfaces that represent the interface boundary between the fluid, struts, and the neighboring tissue. Figure 3 shows the stented core unit cell schematic and geometric data of each stent design.

RESULTS The hemodynamic features of the simplified and patient-

specific stented arterial core geometries are compared.


Differences in WSS, normal (radial) stresses, and vorticity (de-mixing) are reported which confirms the effect of stent design, specifically strut positioning bridge pattern design on hemodynamics. Physiological WSS in an artery is normally within a range from 1 Pa to 7 Pa [26].However, WSS that exceed this range and occur at localizes sites can influence dislodging and/or migration of the stent. Figure 4 present the WSS contour plots of the simplified stented cores for stent

Figure 3: Unit cell geometric data: (a) Stent Design A, (b) Stent Design C, and (c) Stent Design (c).

designs A, B, and C. The highest WSS values were found to occur over the center of the curved region within the surface of the stent struts as shown in Figure 4.

Also Figure 4 shows the regions of low WSS before and after each strut. For the simplified stented core and patient-specific stented core CFD analyses, WSS follows a trend for all stents except for regions between the strut and bridge, as shown in Figure 4. For stent designs A and C, a larger area of arterial tissue is exposed to relatively low WSS (light blue area in Figure 4 between stent struts), compared to Stent Design B.For Stent Design B, the arterial tissue wall area within the first unit cell experiences a relatively high WSS compared to the area around the stent struts. For all three stents, WSS had a high value proximal to the stent struts and a relatively lower value in the area occupied or exposed to neighboring arterial wall tissue.Figure 5 shows the point averaged normalized WSS patterns for all stents averaged over the entire stent wall core impression for the simplified stented cores and the patient-specific arterial stented cores. Figure 5 and 6 shows the scaled WSS and radial stress results of the incoming fluid. The stresses are scaled with the dynamic pressure and the cycle time

is scaled with the total time, t=3 sec. As shown in Figure 5, the maximum normalized WSS for all stents occur at the point of maximum inlet velocity (systole) and the minimum normalized WSS occurs at the minimum inlet velocity (diastole).

Figure 4: WSS contour plots for: (a) Stent Design A, (b)

Stent Design B, and (c) Stent Design C.

(a)

(b) Figure 5: Point averaged stresses vs. normalized time

simplified stented core and patient-specific stented core.

Wall Shear Stress


When comparing the normalized WSS results for the simplified arterial stented core results compared to the patient-specific stented core results, the percent difference of the point averaged normalized WSS for Stent Design A is 44.7%, 44.4% for Stent Design B, 55.8% for Stent Design C, and 25.65% for the Simplified Stent Design. Maximum WSS within the stented region for all stent designs are localized within the curved regions of the struts. Figure 6 shows the normalized point averaged radial stress trend for all stents averaged over the entire stent wall impression for the simplified stented cores and the patient-specific arterial stented cores. When comparing the normalized radial stress results for the patient-specific arterial stented cores compared to the simplified arterial stented cores, the percent difference in results for all stent designs exceed 98.3%. Figure 7 present plots of the distributions of normalized WSS as a function of normalized strut distance within the unit cell of each stent design. The distance is scaled with the total distance from strut to strut. In Figures 7a and 7b, the trend of WSS within the area where the tissue is exposed to the blood has a threshold region of minimum WSS. The threshold of minimum WSS may be attributed to the large area of tissue exposure which minimizes flow disturbance and also minimizes flow resistance.In Figure 7c, the maximum WSS occurs in the unit cell area where the tissue is exposed to blood flow which is due to the minimum area that exists between struts for stent design C.

Also stent design C has more struts compared to stent designs A and B, which influence flow disturbance and WSS. A general conclusion that can be drawn from these analyses is that large arterial tissue exposure within the stent unit cell experiences a lower WSS compared to stents with small arterial tissue exposure. The flow distributions and WSS in larger arterial tissue exposure minimize due to the absence of stent struts. Figure 8 present plots of distributions of normalized radial stress as a function of normalized strut distance within the unit cell of each stent design.The trend of increasing stress with smaller arterial tissue exposure is observed in the results presented in Figure 7.

When comparing the magnitudes of WSS and radial stress, the radial stresses that occur in both the simplified stent arterial cores and patient-specific arterial stented cores are much smaller compared to WSS and can be neglected. The presence of a stent inside the arterial vessel gives rise to flow resistance and flow recirculation (vorticity). Vorticity is defined as the spin or rotation of a fluid. Vorticity is an important parameter in accessing hemodynamic factors because the magnitude of vorticity influences the separation of blood constituents. Vorticity can be defined as the magnitude of de-mixing as it pertains to hemodynamics. Blood is a heterogeneous mixture of various substances and the magnitude of vorticity in stented arteries influences flow alteration and may cause acute pathobiological responses.

Figure 6: Point averaged normalized radial stresses for simplified stented core and patient-specific stented core.

Figure 9 shows plots of the point averaged axial vorticity patterns for all stents averaged over the entire stent wall core impression for the simplified stented cores and the patient-specific arterial stented cores for the entire cardiac cycle. Figure 9a shows that the highest magnitude of flow reversal (de-mixing) exists in Stent Design C, followed by Stent Design A, the Simplified Stent Design, and Stent Design B. Figure 9b shows that the highest magnitude of flow reversal exists in Stent Design A, followed by the Simplified Stent Design, Stent Design C, and lastly Stent Design B. Figures 9a and 9b, show that the hemodynamic behavior of the flow of a stented vessel is not only a function of the stent geometry, but the upstream geometry before the entrance of the stent is important in modeling blood flow in a stented vessel.

Figure 10 shows the magnitude of recirculation that occurs between the struts of all stented cores. The magnitude of recirculation varies with distance from strut to strut. For example, in Figure 10a, Stent Design A has a vorticity magnitude of 650 s-1 which then changes direction and magnitude to 620 s-1 and then minimizes to 198 s-1 before stabilizing to a vorticity magnitude around 90 s-1. Recirculation zones in the cross-flow direction are observed for all stent designs close to the arterial wall.


(a)

(b)

(c) Figure 7: Plot of normalized WSS vs. strut distance. Recirculating flow may be caused by the bridge connectors

that protrude into the space between the struts. It is important to note that the magnitude of vorticity is different for each stent design.Also as shown in Figure 10, the magnitude of vorticity varies with the geometry of the arterial core upstream of the stented region entrance. WSS and radial stresses is a function of stent design, and vorticity is also a function of both stent design and the stented core geometry.

CONCLUSIONS From the results presented in the present work, substantial

differences in hemodynamic parameters were found

(a)

(b)

(c) Figure 8: Plot of normalized radial stress vs. strut

distance. to exist which confirms the effect of stent design on hemodynamics.When comparing strut design, specifically strut positioning, the length in flow direction significantly influences hemodynamic factors. The maximum WSS for all stents occurred at the point of maximum inlet velocity (systole) and the minimum WSS occurred at the minimum inlet velocity (diastole). When comparing WSS results for the simplified arterial stented core results vs. the patient-specific stented core results, the percent difference of the point averaged WSS for Stent Design A was 44.7%, 44.4% for Stent Design B, 55.8% for Stent Design C, and 25.65% for the Simplified Stent Design. When comparing the radial stress results for the patient-specific arterial stented cores compared to the


(a)

(b) Figure 9: Point averaged axial vorticity vs. flow time: (a) axial vorticity for simplified stented cores, and (b) axial

vorticity for patient-specific stented cores. simplified arterial stented cores, the percent difference in results for all stent designs exceeded 98.3%.It is also evident that when accessing hemodynamic factors for simulating blood flow in stented carotid bifurcation arteries, it is important to model the varying arterial cross-section area upstream of the stented core region because this alters hemodynamic results up to 45%.

From the results presented in the present work, the stented patient-specific carotid artery image-based CFD approach is recommended over the commonly used simplified stented core geometry. Unlike previous studies, which have focused on strut thickness on hemodynamics, the present work investigates the difference in hemodynamic forces and flow disturbance arising due to stent pattern design and strut positioning and bridge pattern in both simplified and patient-specific carotid bifurcation arterial geometries.

(a)

(b)

(c) Figure 10: Axial vorticity vs. strut distance.

LIMITATIONS AND FUTURE WORK In clinical practice, a stent is crimped and loaded upon a

balloon that is both tapered and folded onto a catheter. In earlier work found in literature that evaluated the expansion of a stent within normal and stenosed arteries, the artery is generally modeled as an idealized cylinder. In real life situations, the artery does not take on the shape of a cylinder. Future studies will be conducted to introduce the non-axisymmetric nature of the artery and plaque models. Also, a more broad statistical analysis of hemodynamic assessments is needed for stent designs.

ACKNOWLEDGMENTS The present work was funded by the U.S. Department of

Education Title III Grant Program through North Carolina A&T State University.

REFERENCES [1] J. Ricotta, et al., "Cardiovascular disease management: the

0 0.2 0.4 0.6 0.8 1-2

0

2

4

6

8

10

Flow Time (sec)

Axi

al V

ortic

ity (1

/sec

)

Stent Design A, Simplified Stented CoreStent Design B, Simplified Stented CoreStent Design C, Simplified Stented CoreSimplified Stent Design, Simplified Stented Core

0 0.2 0.4 0.6 0.8 1

-10

-5

0

5

10

Flow Time (sec)

Axi

al V

ortic

ity (1

/sec

)

Stent Design A, Arterial Stented CoreStent Design B, Arterial Stented CoreStent Design C, Arterial Stented CoreArterial Stent Design, Arterial Stented Core


need for better diagnostics," Medical & Biological Engineering & Computing, vol. 46, pp. 1059-1068, Nov 2008. [2] S. Chaturvedi, et al., "Carotid endarterectomy - An evidence-based review - Report of the therapeutics and technology assessment subcommittee of the American Academy of Neurology," Neurology, vol. 65, pp. 794-801, Sep 2005. [3] M. A. Creager, et al., "Atherosclerotic Peripheral Vascular Disease Symposium II Executive Summary," Circulation, vol. 118, pp. 2811-2825, Dec 2008. [4] T. L. Forbes, "Preliminary results of Carotid Revascularization Endarterectomy vs Stenting Trial (CREST)," Journal of Vascular Surgery, vol. 51, pp. 1300-1301, May 2010. [5] K. Paraskevas, Mikhailidis, D., and Veith, F., "Mechanisms to explain the poor results of carotid artery stenting (CAS) in symptomatic patients to date and options to improve CAS outcomes," Journal of Vascular Surgery, vol. doi:10/1016/j.jvs.2010.04.019, 2010. [6] P. Mortier, De Beule, M.,Van Loo, D., Verhegghe, B., and Verdonck, P., "Finite element analysis of side branch access during bifurcation stenting," Medical Engineering Physics, vol. 31, pp. 434-440, 2009. [7] I. Pericevic, Lally, C., Toner, D., and Kelly, D., "The influence of plaque composition on underlying arterial wall stress during stent expansion: the case for lesion specific stents," Medical Engineering & Physics, vol. 31, pp. 428-33, 2009. [8] R. Auricchio, and Taylor, R., "Shape-memory alloys: macromodeling and numerical simulations of the superelastic behavior," Computer Methods in Applied Mechanics and Engineering, vol. 146, pp. 281-312, 1997. [9] M. Conti, De Beule, M., Mortier, P., Van Loo, D., Verdonck, P., Vermassen, F., et al., "Nitinol embolic protection filters: design investigation by finite element analysis," Journal of Materials Engineering and Performance, vol. 18, 2009. [10] M. Conti, Auricchio, F., De Beule, M., Verhegghe, B., (06008), "Numerical simulation of Nitiniol peripheral stents: from laser-cutting to deployment in a patient specific anatomy. In proceedings of ESOMAT. 2009., doi:10.1051/esomat/200906008," 2009. [11] D. Martin and F. J. Boyle, "Computational structural modelling of coronary stent deployment: a review," Computer Methods in Biomechanics and Biomedical Engineering, vol. 14, pp. 331-348, 2011. [12] P. Evegren, et al., "Pulsating flow and mass transfer in an asymmetric system of bifurcations," Computers & Fluids, vol. 49, pp. 46-61, Oct 2011. [13] T. H. Zheng, et al., "ASSESSING HEMODYNAMIC PERFORMANCES OF SMALL DIAMETER HELICAL GRAFTS: TRANSIENT SIMULATION," Journal of Mechanics in Medicine and Biology, vol. 12, Mar 2012. [14] Y. M. Hoi, et al., "Effect of Common Carotid Artery Inlet Length on Normal Carotid Bifurcation Hemodynamics," Journal of Biomechanical Engineering-Transactions of the

Asme, vol. 132, Dec 2010. [15] A. Marzo, et al., "Computational Hemodynamics in Cerebral Aneurysms: The Effects of Modeled Versus Measured Boundary Conditions," Annals of Biomedical Engineering, vol. 39, pp. 884-896, Feb 2011. [16] K. R. Moyle, et al., "Inlet conditions for image-based CFD models of the carotid bifurcation: Is it reasonable to assume fully developed flow?," Journal of Biomechanical Engineering-Transactions of the Asme, vol. 128, pp. 371-379, Jun 2006. [17] J. L. Berry, et al., "Experimental and computational flow evaluation of coronary stents," Annals of Biomedical Engineering, vol. 28, pp. 386-398, Apr 2000. [18] D. Rajamohan, et al., "Developing pulsatile flow in a deployed coronary stent," Journal of Biomechanical Engineering-Transactions of the Asme, vol. 128, pp. 347-359, Jun 2006. [19] R. Balossino, et al., "Effects of different stent designs on local hemodynamics in stented arteries," Journal of Biomechanics, vol. 41, pp. 1053-1061, 2008. [20] L. Stockx, "Techniques in Carotid Artery Stenting," European Journal of Radiology, vol. 60, pp. 11-3, 2006. [21] S. Sayeed, Stanzial, S., Wholey, M., and Makaroun, M., "Angiographic lesion characteristics can predict adverse outcomes after carotid artery stenting," Journal of Vascular Surgery, vol. 47, pp. 81-87, 2008. [22] N. Benard, et al., "Computational approach to estimating the effects of blood properties on changes in intra-stent flow," Annals of Biomedical Engineering, vol. 34, pp. 1259-1271, Aug 2006. [23] I. Faik, et al., "Time-dependent 3D simulations of the hemodynamics in a stented coronary artery," Biomedical Materials, vol. 2, pp. S28-S37, Mar 2007. [24] T. Seo, et al., "Computational study of fluid mechanical disturbance induced by endovascular stents," Annals of Biomedical Engineering, vol. 33, pp. 444-456, Apr 2005. [25] J. Hewlin, R.L., and Kizito, J.P., "Evaluation of The Effect of Simplified and Patient-Specific Arterial Geometry On Hemodynamic Flow In Stenosed Carotid Bifurcation Arteries," ASME Early Career Technical Journal, vol. 10, pp. 39-44, 2011. [26] A. M. Malek, Alper, S.L., Izumo, S.,, "Hemodynamic Shear Stress And Its Role In Atherosclerosis," Journal of the Americal Medical Association, vol. 282, pp. 2035-2042, 1999.




MODELING VENTILATION AND PERFUSION MISMATCH

DURING ACUTE LUNG INJURY

Le Yang, Ramana M. Pidaparti Computational Intelligence and Simulation Lab

Department of Mechanical Engineering Virginia Commonwealth University

Richmond, VA, USA

Huan Mo Bioinformatics Program

Virginia Commonwealth University Richmond, VA, USA

ABSTRACT We propose a 2D multi-scale model of

ventilation/perfusion mismatch during acute lung injury (ALI). The inflammation and tissue damage model is based on our previous work, which couples sub-cellular signaling of type II alveolar epithelial cell apoptosis with tissue level events including surfactant production, permeability change and alveolar edema formation. Our model here newly incorporates three aspects of alveolar physiology. First, a hierarchical model of gas exchange is constructed based on the branching structure of one acinus. Second, we incorporate vessel adaptation to oxygen level to the acinus perfusion model. Vessels under hypoxia will constrict while those under normaxia or hyperoxia will dilate. Third, we explicitly model the formation of pulmonary interstitial edema based on Starling’s formula. Both vasogenic edema and permeability edema are considered. The model is able to predict ventilation/perfusion mismatch during a progressing acute lung injury under different internal/external conditions, e.g. the vertical position of the acinus with respect to the upright lung, the mechanical ventilation parameters, etc.

INTRODUCTION Acute lung injury (ALI) occurs due to direct or indirect

insult to the lung. It possesses a mortality rate of 31%-67%. Early deaths from ALI are due to hypoxemia or respiratory failure, while late deaths are caused by sepsis and multi-organ distress syndrome (MODS). Survivors usually develop lung fibrosis. Patients with ALI often require mechanical ventilation to achieve a normal gas exchange function. However, ventilation induced lung injury (VILI) complicates the pathology by volutrauma, atelectrauma or biotrauma mechanisms.

The main pathology of ALI is increased permeability of alveolar-capillary interface and formation of pulmonary edema. Protein-rich fluid leaks out of lung vasculature to interstitium (pulmonary interstitial edema) and/or into alveolar space (alveolar pulmonary edema). Alveolar pulmonary edema inhibits the function of surfactant, as well as provides a nutrient rich environment for pathogen. Pulmonary interstitial edema

reduces lung tissue compliance and interferes with oxygen diffusion, resulting in hypoxemia and respiratory failure.

Previous modeling efforts related to normal lung has covered perfusion [1, 2, 3], ventilation [4, 5, 6], gas exchange [7, 8], etc. A few modeling efforts have tried to resolve the pathology of ALI, including gas exchange [9] and stress induced inflammation [10]. However, these ALI models ignore the hierarchical structure of pulmonary tissue thus cannot reveal the effect of structural geometry on the lung function or pathology. What’s more, they mostly focus on one aspect of pathology while making very simple assumption about other facets.

Our goal in this work is to model pathology development in an acinus of diseased lung in a realistic geometry, and to integrate the crucial aspects of pathology together. The components of our model and the connections between them are shown in Figure 1.

Figure 1. System Layout


METHOD Modeling parallel and serial acinar perfusion. This part of the model is following the work by A.R. Clark et al. [1] They represent the branching structure of acinus by dichotomy of 9 generations. Intra-acinar arterioles and venules branch along the alveolated airways and are connected by capillaries covering the alveoli present in each generation. Capillaries, which have no apparent branching structure, are treated as “sheet”. Modeling tissue damage and repair. We have previously modeled tissue damage and repair during Acute Lung Injury on a 2D plane [11]. In order to model tissue damage and repair on the hierarchical structure of acinus, the branching structure is “flattened” to a 2D square area, with the alveoli in each generation have contacts with those in the previous two generations and next two generations (Figure 2). All alveoli (each represented by one block in Figure 2) contribute exactly the same as in [11] when we calculate the tissue damage and repair, while their contribution to pulmonary edema is calculated generation by generation, averaging across alveoli with the same color (same generation).

Figure 2. Hierarchical structure of acinus flattened to 2D plane. Only generation1-7 is shown. Generations 8 and 9

are arranged similarly.

Modeling pulmonary interstitial edema.

Our model of edema formation is based on Starling’s formula and is formed as a system of mixture of ODE with other equations. Starling’s formula is given as, EVLW = {(SA × LP)((P − P ) − σ × (PI− PI ))} – Flymph

which states that the rate of fluid filtration across capillary membrane is proportionally increasing with the cross-membrane gradient of hydrostatic pressure, while proportionally decreasing with that of osmotic pressure.

Pmv (capillary hydrostatic pressure) is input from the perfusion model. Ppmv (interstitial hydrostatic pressure) is

usually a constant but will become more negative resulting from collapse of alveoli. PImv (plasma osmotic pressure) increases as fluid leaks out of capillaries into interstitium. PIpmv (interstitial oncotic pressure) decreases as fluid leaks into interstitium. Table 1 shows the input variables for the model.

Table1. Variables in edema model

Capillary Interstitium

Lymph

V Fluid volume(ml)

VC VI

Q Protein content(mg)

QC QI Q Protein transport rate(mg/h)

QCI QL

J(V) Fluid transport rate(ml/h)

JCI(VCI) JL(VL)

C Protein concentration (mg/ml)

CC CI

P Hydrostatic pressure(mmHg)

PC PI

π Osmotic pressure(mmHg)

πC πI

The modeled system is as follows: C = QV (i = C, I) π = F(C )(i = C, I)

Plasma: 0.3972 ∗ C − 0.005156 ∗ C + 0.000267 ∗ C

Interstitium: 0.4089 ∗ C − 0.0002403 ∗ C + 0.0004011 ∗ C PC = F(PVV) = 8.33 + 0.544 ∗ PVV(if PVV < 18.27)PVV(else) PI = F(VI) = 0.81 ∗ VI − 8.9(if VI < 11.73)0.6(else) dVIdt = VCI − VL JCI = VCI = K ∗ (PC − PI − σ ∗ (πC − πI)) JL = VL = JL + SL(PI − PI ) ddt QI = QCI − QL QCI = PS ∗ (CC − CI) QL = JL ∗ CI ddt (VI + VC) = 0 ddt (QI + QC) = 0

The system variables are defined in Table 2. The formula for plasma osmotic pressure was adopted

from Ref. [14]. The formulas for pulmonary interstitial osmotic pressure and capillary/interstitial hydrostatic pressures are fitted against experimental data from [15].


Table 2. Parameter values in the edema model

Meaning Value References K Rate coefficient for fluid flow

0.0634 [15] σ Protein reflection coefficient

0.875 [15] PS Permeability times capillaries- interstitium surface area

0.0792 by default

[15]

JL Fluid flow rate of lymph at default state

0.1547 [14]

PI0 Interstitial hydrostatic pressure at default state

-0.51 [14]

SL Conductance of lymph system

0.438 [14] PVV Main venous pressure 15 [14]

Modeling parallel and serial acinar gas exchange. We gather the information from the previous two parts of model, namely, the blood flow through each generation of capillaries and the interstitial edema level of alveoli present in each generation. We assume that under sufficient gas exchange, PO2c will equilibrate with PO2A at the end of inspiration. VI ∗ PO I − VA ∗ PO A = Q(PO C − PO ) PO = PO A = VI ∗ PO I + Q ∗ POVA + Q

However, under interstitial edema, PO2c varies according to different levels of edema. The relationship between PO2c and edema level is interpolated from a previous work by Reynolds et al. [9] and expressed as the following: PO = f(PO A, edema) (1) The saturation level of oxygen in blood is calculated according to Rossing and Cain [12] log(PO ) = 0.371 ∗ log SO1 − SO − 0.48(pH − 7.0)+ 0.019T + 0.92

(2) SO2v is then updated

, 1 = ∗ , ∗ ,, (3)

SO i − 1,2 = SO I, 1 (4) And total content of oxygen is calculated according to Capek and Roy [13]

O = (PO ∗ α) + SO ∗ Hb ∗ Hb (5)

The variables for the acinar gas exchange are defined in Table 3.

Table 3. Variables in acinar gas exchange. Meaning Value Unit Q[i] Blood flow through

capillaries of ith generation.

Input from perfusion model

ml/min

Edema[i] Edema level of ith generation.

Input from edema model, range~(0,1)

arbitrary

VI Inspired ventilation ml

VA Alveolar volumn ul

PO2A Partial pressure of oxygen in alveolus.

107.5 mmHg

PO2a = PO2a[i,j]

Partial pressure of oxygen in j(j=1,2) half of arteriole in ith generation (i=1,2,…9). At first, we assume that PO2a is the same across different generations

40 mmHg

PO2v[i,j] Partial pressure of oxygen in j(j=1,2) half of venule in ith generation (i=1,2,…9).

Output mmHg

PO2c[i] Partial pressure of oxygen at the end of capillary sheet in ith generation (i=1,2,…9).

Output mmHg

RESULTS Pulmonary interstitial edema.

Interstitial edema develops as the permeability of capillary endothelial cells increases in response to inflammation. The edema formation process for a single alveolus is given in Figure 3. After we embed the edema formation model into each alveolus, we are able to predict spatially heterogeneous edema formation in an acinus across time (Figure 4).

Figure 3. Edema development of one alveolus.


Figure 4. Edema development across the acinus. The 1st row shows the tissue damage development across time. In the 2nd row, the colorscale corresponds to the level of

interstitial pulmonary edema, with more intense reds corresponding to higher levels of edema.

Hypoxemia during ALI predicted by the gas exchange model. Oxygen diffusion is disrupted during interstitial edema. We

input the edema level in each generation of the acinus (the data from the area corresponding to that generation), and feed into our hierarchical gas exchange model. Partial pressure of oxygen at the end of the capillaries sheet, thus those in the venules drop significantly from 101.7mm Hg to 60.60mm Hg for generation 9 70 hrs after the onset of acute lung injury, which indicates severe hypoxemia. Partial pressures of oxygen in other generations also drop respectively (Table 4).

Table 4. Partial pressure of oxygen in the venules (PO2v in

mmHg) of different generations (i) of the acinus across time (t). Only data in the first half of the venule is shown. The

perfusion through each generation is given in the 2nd column.

i\t (hr)

Q(*e-3 ml)

0 10 20 30 40 50 60 70

1 4.178 101.3 100.1 99.05 96.61 84.56 78.45 65.90 69.96

2 3.855 101.3 100.1 99.03 96.65 84.83 78.64 65.97 69.82

3 3.481 101.4 100.6 99.57 97.46 86.27 79.82 66.67 69.55

4 3.055 101.7 101.7 100.6 99.00 89.13 82.12 68.02 69.19

5 2.581 101.7 101.7 101.7 100.8 91.19 83.60 68.55 68.28

6 2.074 101.7 101.7 101.7 101.7 93.51 85.01 68.56 67.05

7 1.566 101.7 101.7 101.7 101.7 95.80 95.80 73.34 71.39

8 1.111 101.7 101.7 101.7 101.7 94.38 94.38 74.74 68.54

9 0.959 101.7 101.7 101.7 101.7 90.60 90.60 79.90 60.60

DISCUSSION Previous models of acute lung injury focuses on gas

exchange [9] or stress-induced inflammation [10]. They each represent one aspect of the pathology. This is a first attempt to integrate models governing these aspects together. By using mechanistic modeling methods, our acinar perfusion-ventilation-gas exchange model is able to predict the crucial aspects of pathology during acute lung injury: (1) formation of the interstitial edema (2) perfusion through each generation (3) the hypoxemia due to perfusion/ventilation mismatch.

ACKNOWLEDGEMENT The authors thank the U. S. National Science Foundation for sponsoring the research through a grant CMMI-0969062.

REFERENCES [1] Clark, A.R., Burrowes, K.S., and Tawhai, M. H., 2010, “Contribution of serial and parallel microperfusion to spatial variability in pulmonary inter- and intra-acinar blood flow,” J Appl Physiol, 108, pp. 1116-1126. [2] Clark, A. R., Burrowes, K. S., and Tawhai, M. H., 2011, “The impact of micro-embolism size on haemodynamic changes in the pulmonary micro-circulation,” Resp Physiol & Neuro, 175, pp. 365-374. [3] Burrowes, K. S., Clark, A. R., and Tawhai, M. H., 2011, “Blood flow redistribution and ventilation-perfusion mismatch during embolic pulmonary arterial occlusion,” Pulm Circ, 1(3), pp. 365-376. [4] Graham, M.R., 2005. “Mathematical modeling to centre low tidal volumes following acute lung injury: A study with biologically variable ventilation,” Respiratory Research, 6(64). [5] Adams, A. B., 2005, “Mathematical models of pressure controlled ventilation of oleic acid injured pigs,” Math Med and Biol, 22; pp. 99-112. [6] Bitzen, U., 2006, “Lung mechanics in the aging lung and in acute lung injury”,Ph.D. thesis, Lund University, Sweden. [7] Ben-Tal, A., 2006, “Simplified models for gas exchange in the human lungs,” J Theor Biol, 238, pp. 474-495. [8] Hahn, C. E., Black, A. M., Barton, S. A., and Scott, I., 1993, “Gas exchange in a three-compartment lung model analyzed by forcing sinusoids of N2O,” J Appl Physiol, 75, pp. 1863-1876. [9] Reynolds, A., 2010, “A mathematical model of pulmonary gas exchange under inflammatory stress,” J Theor Biol, 264(2), pp. 161–173. [10] KoomBua, K., 2009, “Multiscale modeling of airway inflammation induced by mechanical ventilation,” Ph.D. thesis, Virginia Commonwealth University, Richmond, VA. [11] Yang, L., and Pidaparti, R. M., 2012, “A multi-scale model of Acute Lung Injury-role of alveolar type II epithelial cell apoptosis,” International Conference on Intelligent Biology and Medicine, Nashwille, TN [12] Rossing, R.G., and Cain, S.M., 1966, “A nomogram relating PO2, pH, temperature and hemoglobin saturation in the dog,” J Appl Physiol, 21(1), pp. 195-201. [13] Capek, J. M., and Roy, R.J. 1988, “Noninvasive measurement of cardiac output using partial CO2 rebreathing,” IEEE Trans on Biomed Engg, 35(9), pp. 653-661. [14] Bert, J. L., Bowen, B. D., and Reed, R. K., 1988, “Microvascular exchange and interstitial volume regulation in the rat: model validation,” Am J Physiol, 254, pp. H384-99. [15] Snashall, P. D., Keyes, S. J., Morgan, B.M., and Chung, K.F., 1982, “Pulmonary interstitial compliance: a function of the osmotic constituents of the interstitium,” J Appl Physiol, 53(2), pp. 324-329.


BIBLIOGRAPHY Wojciechowski, B., ”Starling’s law of the capillaries,” Res Clin Keeper, 19, pp.10-11.


section 1 bio-medical / bioengineering - uab...airborne infection is considered to be a major factor...

Documents